Bompelly Ravi K 201305 PHD

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/317371271
LEAN BLOWOUT AND ITS ROBUST SENSING IN SWIRL COMBUSTORS
Thesis · August 2012

DOI: 10.13140/RG.2.2.30746.85445
CITATIONS READS
3 3,392
All content following this page was uploaded by Ravi K. Bompelly on 06 June 2017.
The user has requested enhancement of the downloaded file.

LEAN BLOWOUT AND ITS ROBUST SENSING
IN SWIRL COMBUSTORS
A Dissertation
Presented to
The Academic Faculty
by
Ravi Kiran Bompelly
In Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy in the
School of Aerospace Engineering

Georgia Institute of Technology
May 2013
LEAN BLOWOUT AND ITS ROBUST SENSING
IN SWIRL COMBUSTORS
Approved by:
Dr. Jerry Seitzman, Advisor Dr. Benn T. Zinn

School of Aerospace Engineering Aerospace Engineering
Georgia Institute of Technology Georgia Institute of Technology
Dr. Timothy Lieuwen Dr. Clarence T. Chang

School of Aerospace Engineering Combustion Branch
Georgia Institute of Technology NASA Glenn Research Center, OH.
Dr. Jeff Jagoda

Georgia Institute of Technology
Date Approved: 22nd August 2012
To my family
For their love and support

ACKNOWLEDGEMENTS
First of all, I would like to thank my advisor Dr. Jerry Seitzman for believing in
me and giving me the opportunity to work on my PhD research. His guidance, support
and constructive criticism helped me greatly in accomplishing the research tasks. It was a
real pleasure working with him and the things I learnt from him will be an invaluable
asset to my career.
I would also like to thank Dr. Tim Lieuwen for his guidance and support during
the initial part of the research. In addition I would like to thank Dr. Jeff Jagoda, Dr. Ben
T. Zinn and Dr. Clarence Chang for being on my thesis committee and providing helpful
comments.
I would like to acknowledge NASA Glenn Research Center for funding this
research project. I thank Mr. John Delaat and Dr. Clarence Chang for monitoring the
project and providing support and guidance.
I would like to thank the research staff at the combustion lab for maintaining the
experimental facilities in great shape and helping me with the usage. I would like to thank
Bobby Noble, David Scarbrough, Sasha Bibik, Eugene Lubarsky, Yedidia Neumeier and
Brad Ochs for their help and advice in conducting my experiments.
I would like to thank all my fellow graduate students at the combustion lab for the
great research and social environment they provided. The intellectual discussions I had
with them helped me immensely in understanding things clearly and advancing my
knowledge. A special thanks goes to Karthik for helping me with running my
experiments at odd times of the day. I would like to thank Yash, Chilu, Yogish, Sai,
iv
Arun, Nishant, Sampath, Jack, Ianko, Ben, Jacheol, Yong, JP and Nori for their help in
my hour of need. I thank my qualifying exam study group, Jackie, Jack, Shreekrishna,
Tom and Caleb, who made the task much easier.
Finally I thank my family for their love and support throughout the journey of my life.
v
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ............................................................................................... iv
LIST OF TABLES ............................................................................................................. ix
LIST OF FIGURES ............................................................................................................ x
NOMENCLATURE ........................................................................................................ xvi
SUMMARY .................................................................................................................... xvii
CHAPTER 1: INTRODUCTION ....................................................................................... 1

1.1. Motivation ........................................................................................................... 1
1.2. Previous Work .................................................................................................... 8
1.3. Overview of Present Work................................................................................ 12
CHAPTER 2: BACKGROUND AND PREVIOUS WORK............................................ 14

2.1. Swirl Stabilized Combustion ............................................................................ 14
2.2. Lean Blowout .................................................................................................... 20
2.3. LBO Margin Sensing: Previous Work .............................................................. 24
2.4. Combustion Dynamics ...................................................................................... 28
2.5. Passive Control of LBO in Aircraft Engines .................................................... 29
CHAPTER 3: EXPERIMENTAL AND MODELING APPROACHES.......................... 32

3.1. Gas-Fueled Combustor Setup ........................................................................... 32
3.1.1. Combustor Design and Flow Facility ....................................................... 32
3.1.2. LBO Sensing and Diagnostics .................................................................. 37
3.2. Liquid-Fueled Combustor Setup ....................................................................... 38
3.2.1. Combustor Design and Flow Facility ....................................................... 38
3.2.2. LBO Sensing and Diagnostics .................................................................. 47
3.3. Turbofan Engine Model and Control ................................................................ 48
3.3.1. High Bypass Turbofan Engine Model ...................................................... 48
3.3.2. Engine Control .......................................................................................... 50
3.3.3. Example Control Results .......................................................................... 53
CHAPTER 4: LBO MARGIN SENSING: OPTICAL DETECTION .............................. 55

4.1. LBO Margin Sensing in the Gas-Fueled Combustor ........................................ 55
vi
4.1.1. Instability Characteristics.......................................................................... 56
4.1.2. LBO Precursor Events .............................................................................. 58
4.1.3. Precursor event detection .......................................................................... 65
4.1.4. LBO Margin Detection ............................................................................. 70
4.2. LBO Margin Sensing in the LDI Combustor .................................................... 76
4.2.1. Instability Characteristics.......................................................................... 77
4.2.2. Precursor events ........................................................................................ 78
4.2.3. LBO Proximity Sensing ............................................................................ 81
4.2.4. CH* vs OH* Signals................................................................................... 86
4.3. Summary ........................................................................................................... 88
CHAPTER 5: LBO MARGIN SENSING: ACOUSTIC DETECTION .......................... 90

5.1. LBO Margin Sensing in the Gas-Fueled Combustor ........................................ 90
5.2. LBO Margin Sensing in the LDI Combustor .................................................... 94
5.3. Effect of Instability and Precursors Timescales................................................ 96
5.4. Effect of Reflections on Acoustic Precursor Detection .................................... 99
5.4.1. Acoustic Model ......................................................................................... 99
5.4.2. Model Results for LDI Combustor ......................................................... 102
5.4.3. Model Results for Gas-Fueled Combustor ............................................. 105
5.5. Summary ......................................................................................................... 107
CHAPTER 6:LBO MARGIN SENSING: RAPID TRANSIENTS................................ 108

6.1. Stochastic Model for Precursor Event Occurrence ......................................... 108
6.2. Probability of Precursor Event Occurrence in a Rapid Transient ................... 113
6.3. Precursor Event Simulation in Real-time ....................................................... 117
6.4. Time Response Improvement in Engine Decelerations with LBO Margin
Sensing ...................................................................................................................... 119
6.4.1. Transient Response Improvement........................................................... 120
6.4.2. Event based active control ...................................................................... 124
6.5. Summary ......................................................................................................... 127
CHAPTER 7: CONCLUSIONS AND RECOMMENDATIONS .................................. 129

7.1 Summary and Conclusions ............................................................................. 129
7.1.1. Robust LBO Margin Sensing .................................................................. 130
7.1.2. LBO Margin Sensing in Rapid Transients .............................................. 133
7.2. Recommendations for Future Work................................................................ 134
vii
APPENDIX A: ENGINE FUEL-AIR RATIO ESTIMATION ...................................... 138
APPENDIX B: LDI FUEL NOZZLE TESTING ........................................................... 140
APPENDIX C: LDI COMBUSTOR DEVELOPMENT ................................................ 144
REFERENCES ............................................................................................................... 148
viii
LIST OF TABLES
Table 1. Design point specifications of the turbofan engine. ........................................... 50
Table 2. Engine transfer function parameters at full power and idle ................................ 52
Table 3. Kolmogorov-Smirnov hypothesis test results. .................................................. 111
Table 4. Event simulation algorithm for non-stationary Poisson process. ..................... 118
ix
LIST OF FIGURES
Figure 1. NOx and CO emissions in a Dry Low NOx (DLN) combustion system as a
function of combustion product temperature. [7] ............................................................... 4
Figure 2. Variation with engine power of pilot and main primary zone equivalence ratios
of a lean-operation aero engine combustor..........................................................................6
Figure 3. Variation of pilot primary zone equivalence ratio during a rapid transient from
full power to idle..................................................................................................................7
Figure 4. Illustration of a swirling flow field structure. .................................................... 18
Figure 5. Illustrative flame configurations in premixed swirl combustors. ...................... 19
Figure 6. (a) Stable flame during nominal operation (b) Sequence of flame images during
an extinction and re-ignition event; image separation 4 ms. (Ref [69]) ........................... 24
Figure 7. Precursor events in the OH* optical signal along with thresholds. (Ref. [69]) 25
Figure 8. Variation of average number of precursor events with equivalence ratio. (Ref.
[69])................................................................................................................................... 26
Figure 9. Precursor events in the acoustic signal. (Ref. [71]) .......................................... 27
Figure 10. Power spectra of acoustic signals in a premixed swirl combustor. (Ref. [71]) 27
Figure 11. Response of an LBO control system to varying operating conditions. (Ref.
[69])................................................................................................................................... 28
Figure 12. Illustrative engine control schedules during transients. (Ref. [75]) ................ 31
Figure 13. Schematic diagram of the gas fueled combustor setup....................................33
Figure 14. Fuel injection orifice module........................................................................... 34
Figure 15. Schematic diagram of fuel and air supply, control and monitoring setups. .... 35
Figure 16. Flame configurations in the combustor for different equivalence ratios at a
cold flow velocity of 2.5 m/s: (a) Φ=0.88, (b) Φ=0.79, (c) Φ=0.66. ................................ 36
Figure 17. NASA 9 element LDI injector. [5] ................................................................. 39
Figure 18. (a) LDI element assembly with venturi, swirler and fuel nozzle. (b) swirler. . 39
Figure 19. (a) Cross sectional view of converging-diverging venturi section (all
dimensions in mm) and (b) fuel nozzle stem cross sectional view. .................................. 40
x
Figure 20. Perforated plate. ............................................................................................... 41
Figure 21. Cross sectional view of the single element LDI combustor. ........................... 42
Figure 22. LDI combustor in the high pressure test rig. ................................................... 43
Figure 23. Schematic diagram of fuel supply and control setup....................................... 44
Figure 24. Pressure dependence of fuel nozzle mass flow rate for Jet-A. ........................ 45
Figure 25. Flame in the LDI combustor at (a) p=1 atm, v=14 m/s, T= 663 K ,
Φoverall=0.41; (b) p=2 atm, v=10.6 m/s, T=682K, Φoverall=0.28; and (c) p=4 atm,
v=12 m/s, T=733 K, Φoverall=0.23 . ................................................................................... 47
Figure 26. Schematic of the simulated turbofan engine with flow paths...........................49
Figure 27. Architecture of the engine control system. ...................................................... 51
Figure 28. An example full power to idle transient starting at t=1 sec. ............................ 54
Figure 29. Acoustic signal time traces for combustor operation near LBO, at Φ=0.71 for
instability w/o Φ′ Φ=0.72 for instability w/ Φ′. ................................................................ 57
Figure 30. Power spectra of acoustic signals for combustor operation near LBO at
Φ=0.71 for instability w/o Φ′ Φ=0.72 for instability w/ Φ′. ............................................. 57
Figure 31. OH* chemiluminescence signal time trace with a precursor event occurring
between 100-150 ms for dynamically stable operationat Φ=0.704. ................................. 59
Figure 32. OH* signal time traces with precursor events for dynamically unstable
operation w/o Φ' and w/ Φ'. Mean equivalence ratios: 0.698 (w/o Φ'), 0.714 (w/ Φ')..... 60
Figure 33. High speed flame chemiluminescence images for instability w/o Φ', Φ =0.698.
(Top row) Images during normal combustion over an instability cycle with an image
separation of 0.526 ms. (Bottom row) Images during a precursor event, every 13th image
shown (effective image separation of ~6.8 ms). In the color bar red, represents the highest
chemiluminescence emission intensity and blue represents the lowest. ........................... 61
Figure 34. High speed flame chemiluminescence images for instability w/ Φ', average Φ
=0.711. (Top row) Images during normal combustor operation with the first four images
during a high amplitude instability cycle, separated by 1 ms; last two frames during a
period of low instability. (Bottom row) Images during a precursor event with an image
separation of 6 ms. ............................................................................................................ 63
Figure 35. Optical signal time trace acquired simultaneously with the high-speed images
in Figure 34 (top row images), for instability w/ Φ'. Image locations are indicated by
circles.................................................................................................................................64
xi
Figure 36. An example precursor event in optical signal along with the thresholds used
for its detection. ................................................................................................................ 65
Figure 37. Amplitude spectra of optical signals for instability w/ Φ' and w/o Φ'.
Equivalence ratios: 0.698 (w/o Φ'), 0.714 (w/ Φ'). ........................................................... 67
Figure 38. Precursor events in low-pass filtered (50 Hz) optical signals for dynamically
stable and unstable w/o Φ' and w/ Φ' conditions. ............................................................. 68
Figure 39. PDFs of low-pass filtered optical signals along with corresponding Gaussian
PDFs for far from LBO and near LBO operation for instability w/o Φ'. An inset plot with
reduced X and Y scales for near LBO case is also
shown.................................................................................................................................69
Figure 40. Average number of events per second for different lower threshold settings for
instability w/o Φ'. Upper threshold=-1 and minimum time constraint=10 ms............. 71
Figure 41. Average event occurrence rates obtained from low-pass filtered optical signals
for dynamically stable, dynamically unstable w/o Φ' and w/ Φ'. ...................................... 73
Figure 42. Average event duration obtained from un-filtered optical signals for
dynamically stable, dynamically unstable w/o Φ' and w/ Φ'. ........................................... 74
Figure 43. Average event rate and Stability Index (SI) for instability w/o Φ' obtained
from low-pass filtered optical signals. Left vertical axis scale: event rate and SI
normalized with average event rate SI in the range Φ-ΦLBO = 0.03-0.05. Right axis scale:
actual scale for SI. ............................................................................................................. 75
Figure 44. Average event rate and Stability Index (SI) for instability w/ Φ' obtained from
low-pass filtered optical signals. Left axis scale: event rate and SI normalized with
average event rate SI in the range Φ-Φ LBO = 0.023-0.044. ........................................... 76
Figure 45. Power spectra of CH* signal and acoustic signals for LDI combustor
operation near LBO at Φ/ΦLBO = 1.09; combustor pressure = 2 atm. ............................... 77
Figure 46. CH* signal time trace with precursor event features indicated by arrows for
combustor operation close to LBO at Φ/ΦLBO = 1.02. ...................................................... 78
Figure 47. 500 Hz low-pass filtered CH* signal with precursors indicated by arrows for
combustor operation close to LBO at Φ/ΦLBO = 1.02. ...................................................... 80
Figure 48. High speed flame images during a precursor event in the LDI combustor at 2
atm operation near LBO.....................................................................................................81
Figure 49. Average event rates for different lower threshold settings; upper threshold=-
0.5 and minimum time constraint =1.4 ms.............................................................. ....... 82
xii
Figure 50. Average event rates for different minimum duration constraints; lower
threshold=-3.5 and upper threshold=-0.5. .............................................................. 83
Figure 51. Average event rates for 2 and 4 atm operations. ............................................. 83
Figure 52. Average duration and normalized modulation depths of an event. ................. 85
Figure 53. Average event rate and Stability Index (SI) comparison. Left axis scale: event
rate and SI normalized with average values in Φ/ ΦLBO range: 1.25-1.47. ....................... 85
Figure 54. (top) Normalized equivalence ratio variation with time (bottom) Number of
events and SI calculated in moving 1 s time windows...................................................... 86
Figure 55. Average event rate from low-pass filtered CH* and OH* signals with same
threshold settings. ............................................................................................................. 87
Figure 56. CH* and OH* signals during a precursor detected in the filtered OH* signal;
(top) unfiltered (bottom) low-pass filtered signals. (Φ/ ΦLBO=1.54). ............................... 88
Figure 57. (Top) Precursor event in the OH* signal for dynamically stable operation with
a precursor event at ~50-90 ms; (bottom) simultaneous acoustic signal. ......................... 91
Figure 58. Acoustic signals during precursor events observed in optical signals (Figure
32) for instability w/o Φ' and w/ Φ'. ................................................................................. 91
Figure 59. Low-pass filtered acoustic signal with precursor events for dynamically stable
and unstable w/o Φ' and w/ Φ'. ......................................................................................... 93
Figure 60. Average event rates obtained from low-pass filtered acoustic signals for
dynamically stable and unstable conditions. ..................................................................... 94
Figure 61. (Top) Raw acoustic signal time trace from the LDI combustor during
precursor events observed in optical signals, as marked by boxes; (bottom) low-pass
filtered signal. ................................................................................................................... 95
Figure 62. Precursor events and their spectral content: (a) optical precursor, (b) its
amplitude spectrum, (c) acoustic precursor, and (d) its amplitude spectrum. .................. 97
Figure 63. Model optical and acoustic signals with instability and precursors having
similar time scales: (a) optical instability signal with a precursor; (b) acoustic instability
signal with a precursor; (c) low-pass filtered optical signal; and (d) low-pass filtered
acoustic signal. .................................................................................................................. 98
Figure 64. Schematic of acoustic model system used to investigate the influence of
reflections on the detected acoustic pressure signal. The flow inlet is at the left; two
possible transducer locations are indicated. .................................................................... 100
xiii
Figure 65. An acoustic precursor reaching the transducer without any reflections (initial
precursor) and the resultant of multiple reflections (reflected precursor). ..................... 103
Figure 66. Amplitude spectra of the initial and reflected precursors. ............................. 103
Figure 67. Model instability signals with initial and reflected precursors added to them:
(a) instability signal with initial precursor; (b) instability signal with reflected precursor;
(c) low-pass filtered version of (a); (d) low-pass filtered version of
(b).....................................................................................................................................104
Figure 68. Simulated reflected precursor signals with the initial precursor having different
durations in the LDI combustor. ..................................................................................... 105
Figure 69. Simulated reflected precursor signals with the initial precursors having
different durations in the gas-fueled combustor. ............................................................ 106
Figure 70. CDFs of measured time between successive events (TBE) for the LDI
combustor (symbols) along with corresponding exponential CDFs (lines).................... 110
Figure 71. Empirical CDFs of the normalized time between events, with events detected
using a threshold of µ-3. An exponential CDF with a unit mean time between events is
also shown....................................................................................................................... 111
Figure 72. Empirical CDFs of time between events for the gas fueled combustor, having
instability without equivalence ratio oscillations, along with corresponding exponential
CDFs. .............................................................................................................................. 113
Figure 73. Probability of at least one or two events occurring for a linear transient in 
from 1.20 to 1, as a function of the transient duration. Results are based on the LDI
combustor event rates. .................................................................................................... 116
Figure 74. Probability of at least one or two events occurring for a linear transient in 
from 1.06 to 1, as a function of the transient duration. Results are based on the gas-fueled
combustor event rates. .................................................................................................... 116
Figure 75. Experimental and simulated number of events in a one second interval for a
transient in equivalence ratio in the LDI combustor. ...................................................... 118
Figure 76. (a) Gaussian PDF of LBO equivalence ratio; (b) probability of LBO
equivalence ratio being above a given equivalence ratio................................................ 121
Figure 77. Engine Ratio Unit (RU) during a full power to idle transient for nominal and
modified RU limit control cases. .................................................................................... 122
Figure 78. Thrust response for full power to idle transient at sea level static conditions
with the nominal RU limiter and the modified RU limiter. ............................................ 122
xiv
Figure 79. (a) Average event rates for different lower threshold settings; (b) probability of
failure to sense LBO for different event rate trends........................................................124
Figure 80. Equivalence ratio response for the modified RU limiter case and event based
control case assuming ΦLBO=0.51. .................................................................................. 126
Figure 81. Simulated events in real-time (top) and normalized equivalence ratio (bottom)
for event based control. ................................................................................................... 126
Figure 82. Thrust responses for the modified RU limiter case and with event based
control case. .................................................................................................................... 127
Figure 83. An example flame configuration during precursor events from a time resolved
chemiluminescence image. ............................................................................................. 135
Figure 84. The original bent fuel nozzle and the modified straight nozzle. .................. 140
Figure 85. Spray images for (left) bent nozzle (right) straight nozzle at 70 psi pressure
differential. ...................................................................................................................... 141
Figure 86. Droplet sauter mean drop diameters (left) at an axial location of 2cm (right)
at an axial location of 5cm, from nozzle tip, for bent and straight nozzles. ................... 142
Figure 87. Droplet average axial velocities (left) at an axial location of 2cm (right) at an
axial location of 5cm, from nozzle tip, for bent and straight nozzles. ............................ 143
Figure 88. Flame in the LDI combustor for (a) P=1.17 atm, T=663 K, V=14 m/s,
Φoverall=0.5 (b) P=2atm, T=682K, V=10.6m/s, Φoverall =0.3. .......................................... 146
Figure 89. Flame in the combustor at P=1.13 atm, T=654K, V= 13.4m/s, Φoverall=0.32, for
fuel nozzle position upstream of venturi throat. ............................................................. 146
Figure 90. Flame in the combustor at (a) P=1atm, T=663K , V=14 m/s, Φ overall=0.41 (b)
P=2atm, T=682K, V=10.6m/s, Φoverall=0.28 (c) P=4atm, T=733K, V=12m/s,
Φoverall=0.23 . ................................................................................................................... 147
xv
NOMENCLATURE
a Duct diameter
C Speed of sound
k Acoustic wave number
M Mach number
Ps 3 Combustor inlet static pressure
R Reflection coefficient
T Transmission coefficient
U Uniform random number
Wf Fuel flow rate
Z Acoustic impedance
 Average event occurrence rate
 Expected number of events in a transient
 Signal mean
 Normalized equivalence ratio
 Equivalence ratio
 Density
 Signal standard deviation
LBO Lean Blow Out
LDI Lean Direct Injection
RU Ratio Unit
SI Stability Index
TBE Time Between Events
xvi
SUMMARY
Lean combustion is increasingly employed in both ground-based gas turbines and
aircraft engines for minimizing NOx emissions. Operating under lean conditions increases
the risk of Lean Blowout (LBO). Thus LBO proximity sensors, combined with
appropriate blowout prevention systems, have the potential to improve the performance
of engines. In previous studies, atmospheric pressure, swirl flames near LBO have been
observed to exhibit partial extinction and re-ignition events called LBO precursors.
Detecting these precursor events in optical and acoustic signals with simple non-intrusive
sensors provided a measure of LBO proximity.
This thesis examines robust LBO margin sensing approaches, by exploring LBO
precursors in the presence of combustion dynamics and for combustor operating
conditions that are more representative of practical combustors, i.e., elevated pressure
and preheat temperature. To this end, two combustors were used: a gas-fueled,
atmospheric pressure combustor that exhibits pronounced combustion dynamics under a
wide range of lean conditions, and a low NOx, liquid-fueled Lean Direct Injection (LDI)
combustor, operating at elevated pressure and preheat temperature. In the gas-fueled
combustor, flame extinction and re-ignition LBO precursor events were observed in the
presence of strong combustion dynamics, and were similar to those observed in
dynamically stable conditions. However, the signature of the events in the raw optical
signals have different characteristics under various operating conditions. Low-pass
filtering and a single threshold-based event detection algorithm provided robust precursor
sensing, regardless of the type or level of dynamic instability. The same algorithm
provides robust event detection in the LDI combustor, which also exhibits low level
dynamic oscillations. Compared to the gas-fueled combustor, the LDI events have
weaker signatures, much shorter durations, but considerably higher occurrence rates. The
xvii
disparity in precursor durations is due to a flame mode switch that occurs during
precursors in the gas-fueled combustor, which is absent in the LDI combustor.
Acoustic sensing was also investigated in both the combustors. Low-pass filtering
is required to reveal a precursor signature under dynamically unstable conditions in the
gas-fueled combustor. On the other hand in the LDI combustor, neither the raw signals
nor the low-pass filtered signals reveal precursor events. The failure of acoustic sensing is
attributed in part to the lower heat release variations, and the similarity in time scales for
the precursors and dynamic oscillations in the LDI combustor. In addition, the impact of
acoustic reflections from combustor boundaries and transducer placement was addressed
by modeling reflections in a one-dimensional combustor geometry with an impedance
jump caused by the flame.
Implementing LBO margin sensors in gas turbine engines can potentially improve
time response during deceleration transients by allowing lower operating margins.
Occurrence of precursor events under transient operating conditions was examined with a
statistical approach. For example, the rate at which the fuel-air ratio can be safely reduced
might be limited by the requirement that at least one precursor occurs before blowout.
The statistics governing the probability of a precursor event occurring during some time
interval was shown to be reasonably modeled by Poisson statistics. A method has been
developed to select a lower operating margin when LBO proximity sensors are employed,
such that the lowered margin case provides a similar reliability in preventing LBO as the
standard approach utilizing a more restrictive operating margin. Illustrative
improvements in transient response and reliabilities in preventing LBO are presented for
a model turbofan engine. In addition, an event-based, active LBO control approach for
deceleration transients is also demonstrated in the engine simulation.
xviii
CHAPTER 1
INTRODUCTION
1.1. Motivation
Reduction of pollutant emissions from ground-based gas turbine engines and
aircraft engines is essential for protecting air quality and preventing damage to the
environment. Nitrogen oxides (NOx) are a major pollutant with many adverse effects on
the environment. At ground level and low altitudes, NOx emissions contribute to
formation of photochemical smog, harmful ozone and acid rain [1]. NOx emissions from
current subsonic commercial aircraft operating at cruise altitudes (9-13 km) increase
ozone levels along the traffic routes, which could alter the climate [2] . At high altitudes
(17-20 km) in the stratosphere, corresponding to the cruise altitudes of future supersonic
passenger aircraft, NOx emissions would lead to depletion of the ultraviolet-blocking
ozone layer [2].
NOx emissions in lean conditions are generally an exponential function of flame
temperature. Hence lowering average flame temperatures and avoiding local temperature
peaks can greatly reduce NOX emissions. In addition, lowering the residence time
combustion products spend at high temperature (before temperature drops due to work
extraction or heat transfer) can aid in reducing NOx. Fuel lean operation with uniform
fuel-air mixing produces the homogenous and low combustion zone temperatures
required for lowering NOx. Hence, recent gas turbine combustor design approaches have
mainly focused on leaner combustion, using premixed or partially premixed operation as
a preferred option for NOx reduction. Lean premixed operation has reduced NOx
emissions substantially in land based gas turbine engines [3]. Similarly, lean premixed
pre-vaporized (LPP) [4], and Lean Direct Injection (LDI) [5] combustion approaches are
suggested solutions for NOx reduction in aero engines. An alternate approach, Rich-
1
Quench-Lean (RQL) combustion, has also been pursued [6]. While RQL is preferred for
its enhanced stability, incomplete mixing between rich products and secondary air and
higher particulate emissions in RQL favor lean combustion approaches [7].
Continuous combustion in gas turbines, in contrast to intermittent combustion in
reciprocating engines, requires a stable flame to continuously burn fuel. The flame needs
to be stable in high velocity streams employed for producing high heat release rates in
compact volumes. Gas turbine engine main combustors predominantly use flow
recirculation generated by swirl and sudden area expansion for flame stabilization. In
addition, a pilot flame is occasionally employed. The recirculation region provides a
continuous supply of hot products and radicals to the oncoming reactants and helps
stabilize the flame [8]. In addition, flow recirculation creates low velocity regions where
flame speed can match flow speed for flame stabilization. By making a fuel-air mixture
leaner, the stabilization process weakens due to multiple reasons. For example, lean
mixtures result in low product temperatures and radical concentrations in recirculation
zone reducing their ability to ignite reactants. In addition, flame speed decreases for
leaner mixtures, such that it cannot match flow speed, required for stabilization.
Furthermore, lean flames are less able to withstand high flame stretch rates [9], as the
extinction strain rate is lower for lean mixtures. Swirl stabilized flames usually
experience high stretch rates due to high velocity gradients and turbulence levels, making
lean flames more susceptible to extinction. Therefore for sufficiently lean mixtures, a
flame cannot be stabilized, resulting in flame Lean Blowout (LBO). This is also referred
to as static instability of a flame.
Lean blowout results in disruption of essential power or thrust output from
engines and requires a complex relight procedure to restore power. In land-based gas
turbines used for electric power generation, power outage resulting from LBO would
require operators to pay penalties making LBO an expensive problem. In aero engines,
2
LBO is a flight safety hazard. For example, LBO near ground level could lead to a
catastrophic crash since aircrafts would descend rapidly with no thrust and little time for
engine restart. LBO at cruise altitudes can require considerable reduction in altitude as it
is hard to relight at high altitudes where ambient pressure and temperature are low.
The exact conditions at which LBO occurs are hard to predict. For example, LBO
occurrence is dependent on local flow conditions, including turbulence levels,
temperature, equivalence ratio, spray properties (for liquid fuels), fuel composition and
product entrainment into the reactants. These conditions are not precisely known or
predictable during engine operation, and can vary significantly with engine operating
conditions or due to aging effects. In addition, inherent disturbances in engine operating
conditions can push the combustor to LBO. Thus combustors are typically designed with
large operating margins to avoid LBO, i.e., with a flame zone equivalence ratio much
higher than the actual LBO limit. However, excessive margins mean the engine is likely
operating at sub-optimal conditions. If however, LBO margin sensors and control
systems were available to avoid LBO, the required margins could be reduced, and further
reduction of NOx could be achieved.
For land-based gas turbine engines, the operating point may be chosen primarily
to minimize NOx while maintaining allowable CO emissions and preventing LBO and
excessive combustion dynamics. NOx and CO emissions, along with an LBO boundary,
are illustrated in Figure 1 for a “typical” Dry Low NOx (DLN) combustor. Keeping NOx
and CO within the allowable limits (for example, 15 ppm for NOx and 25 ppm for CO)
results in a relatively narrow operating range for combustors. Typically, the minimum
equivalence ratio for preventing excessive CO emissions is higher than the LBO
equivalence ratio. However, this lower bound on equivalence ratio is close to the LBO
limit. DLN combustors often employ radial staging using multiple burners within a
combustor can to provide sufficient turndown. The burner arrangement requires fuel split
3
schedules as a function of engine load, for maintaining low emissions throughout the load
range. For part load operation, some of the burners operate below their LBO limit while
keeping other burners richer, to stabilize the lean burners. In practice, however, it has
been observed that power trips due to LBO occurred due to insufficient stabilization
provided by rich burners [10]. In land-based engines, fuel tuning is performed regularly
in the field, mainly to prevent dynamics. Errors associated with tuning, changes in fuel
properties, variation in ambient conditions can result in occasional LBO trips. LBO
proximity detection systems could be used to avoid these events and improve the
performance of engines.
120 30
NOx
LBO
100 25
Operating
NOx Emissions (ppmvd)

CO Emissions (ppmvd)
range
80 20
60 15
40 10
20
CO 5
0
1500 1600 1700 1800 1900 2000 2100
Firing Temperature (K)
Figure 1. NOx and CO emissions in a Dry Low NOx (DLN) combustion system as
a function of combustion product temperature. [7]
For aero engines, NOx is regulated at low altitudes (take off, approach and
landing). Though not currently regulated at cruise altitudes, NOx at cruise may have the
greatest environmental impact, as engines operate for the most time at cruise. For
4
achieving low emissions using lean combustion, radial staging using a dual annular
combustor [11] or staging within swirlers using multiple swirlers [12] are often employed
to provide sufficient turndown. The combustor configuration can also consist of a pilot
and a main region. The pilot operates alone at low power levels, and at high power levels,
it acts to stabilize the leaner main section. An illustration of average primary zone (a
region without secondary air dilution) equivalence ratios of the pilot and main regions are
shown in Figure 2. In addition, an example LBO margin and LBO limit are shown. The
LBO limit can be expected to decrease with engine power, as higher compressor exit
pressures and temperatures would tend to lower the equivalence ratio where LBO occurs.
For minimizing NOx at the operating points of interest, e.g., cruise or full power take-off,
the design points should be at the lowest possible equivalence ratio. However this can
result in other operating points, e.g., idle or where the main region switches, falling
below the required LBO safety margin. Hence the constraint of having minimum required
LBO margin over the entire operating range of an engine results in non-optimal emission
performance at points where NOx has to be reduced. An LBO control system that ensures
stable engine operation at these off-design conditions would allow reduced LBO margins
and therefore lower NOx emissions at the design points. At the off-design conditions, the
LBO control system would likely be free to act without the constraint to minimize NOx
emissions.
In aero engines, LBO can occur during steady-state operation or during power
reduction transients, i.e., deceleration. To decrease power, fuel flow to a combustor is
reduced, resulting in lower gas temperatures/velocities entering the turbine and reduced
turbine work. This slows down the shaft rotational speed as the turbine cannot provide
sufficient torque to drive the compressor at the same speed. The reduced rate depends on
the inertias of the compressor, turbine and connecting shafts, and on the torque deficit on
the compressor. The mass flow rate of air through the compressor depends on the
compressor shaft speed for a given flight Mach number. While fuel flow can be reduced
5
quickly to reduce power, air flow rate drops rather slowly. This results in lower
combustor equivalence ratios, which push the combustor closer to LBO. This process is
illustrated in Figure 3, where pilot primary zone equivalence ratio is plotted as a function
of engine power during steady-state operation and a fast deceleration transient from full
power to idle. For illustrative purpose, only the pilot zone equivalence ratio is shown, as
the pilot is the main stabilization mechanism.
Pilot Only Pilot + Main

Primary Zone Equivalence Ratio
Pilot
Main
1.0 TO
Cruise
Idle
0.5
LBO Margin
LBO limit
Engine Power
Figure 2. Variation with engine power of pilot and main primary zone
equivalence ratios of a lean-operation aero engine combustor.
Due to the uncertainty in LBO conditions, the minimum allowed equivalence ratio
during a deceleration transient is kept well above the LBO boundary, i.e., a large LBO
margin is used (20% above the LBO limit in [13]). This is achieved by a slower fuel flow
rate drop or raising the minimum allowed equivalence ratio limit in the engine controller.
In either case, the result is a slower engine transient response. With LBO margin
detection sensors, LBO margins can be decreased, improving deceleration transient
response. The improved deceleration response can have multiple benefits, for example,
6
faster aircraft descent or enabling engines to be used more effectively as backup flight
control in case control surfaces, such as the rudder, fail.
Pilot Primary Zone Equivalence Ratio

Pilot Only Pilot + Main
TO
1.0 SS operating
line
Cruise
Idle
0.5
LBO Margin
LBO limit
Engine Power
Figure 3. Variation of pilot primary zone equivalence ratio during a rapid transient
from full power to idle.
In land-based gas turbines used for electric power generation, generator shaft
rotational speed has to be kept nearly constant in order to maintain the electrical
frequency, e.g., 50 or 60 Hz. When load on the generator is rapidly shed, shaft rotational
speed increases, causing an increase in the frequency or excessive turbine speed. A
higher electrical frequency can result in a power trip, whereas excessive turbine speed
can result in mechanical damage to the gas turbine. In both cases, shaft speed has to be
reduced rapidly to ensure safe operation. Similar to the aircraft engine decelerations
described above, shaft speed is reduced by lowering fuel flow, thus pushing combustors
towards LBO. By employing lower LBO margins, fuel flow rate can be reduced further
and shaft speeds can be reduced faster, thus preventing power trips and avoiding engine
damage.
7
From these examples, it is clear that having an LBO proximity sensor would
improve performance and reliability of gas turbine engines. It would permit reduction in
minimum required LBO margins, allowing for reduced NOx emissions and improved
transient response. Ideally, the sensor would be simple in construction, non-intrusive and
capable of working in harsh engine conditions. In addition, the sensor should have good
sensitivity, time response and be robust to varying operating conditions. Therefore the
main objective of this study is to develop a robust sensing methodology capable of
warning of imminent LBO in real time.
1.2. Previous Work
For reliable flame blowout prevention, it is essential to understand the physical
process governing the blowout phenomenon. Such understanding may aid in developing
sensors for blowout prevention. In combustors, blowout can be caused by various fluid
mechanic and chemical process depending on the combustor configuration, flame
stabilization method, and other operating conditions. Generally combustion cannot be
sustained below an extinction limit, where heat release during the residence time of the
gases in the combustor is not sufficient enough to increase the reactant energy
(temperature) above the minimum activation energy required for combustion. Well-
stirred reactor models employing this property sometimes successfully predict blowout
conditions [14]. Most turbine engine combustors use flow recirculation as a primary
method for flame stabilization. Blowout occurs when the flame stabilization mechanism
is not adequate to hold a flame. For example, flow turbulence may induce unsteady
flame stretch sufficient enough to cause local extinctions for lean flames. These
extinctions may result in decreased temperature/radical concentration in the recirculation
zone and ultimately cause blowout [15]. In addition, LBO could occur due to insufficient
time for ignition of reactants in a high velocity stream. High velocities of reactants in the
shear layer between the recirculation zone and free stream would result in small residence
8
times compared to chemical time scales making the reactants hard to ignite. This
phenomenon has been employed in bluff body stabilized flames using Damkohler
number for LBO prediction [16]. Though there is a considerable amount of literature
regarding blow off in bluff body stabilized flames, in swirl stabilized flames there is only
a limited amount of previous work.
Combustor blowout is often observed to be non-abrupt and preceded by transient
unsteady phenomenon. For example in swirl flames near blow out, some unsteady
behavior is often observed in terms of flame area, flame shape or heat release. Hedman et
al. [17] observed oscillations near blowout between a flame that is attached around the
recirculation zone and one that is lifted off from the recirculation zone. In addition, they
observed the oscillating flame condition to be stable for a high swirl case, but it would
occasionally extinguish for medium swirl conditions. Griebel et al. [18] observed
oscillations in flame postion, shape and length oscillating prior to blowout. Similar
unsteady flame behavior was observed in a non-premixed, swirl stabilized combustor by
Sturgess et al. [19, 20]. They observed flame liftoff and subsequently sever intermittency
with equivalence ratio reduction. Further reduction in equivalence ratio resulted in large
scale axial flame movement before blowing out. The severe unsteady nature of the flame
indicates the complexity of the blow out process, making it hard to capture by modeling.
The unsteady behavior occurring with a sufficient margin, e.g., in equivalence ratio,
above the blowout limit can provide a means for real-time prediction of approaching
LBO.
Based on this unsteady behavior prior to LBO, Thiruchengode et al. [21] sensed
approaching lean blowout in a premixed, gas-fueled swirl combustor operating at
atmospheric pressure. The unsteady behavior was found to be associated with partial
flame extinction around the inner recirculation zone followed by its re-ignition. The
extinction and re-ignition events were called LBO precursor events as they precede LBO
9
and ultimately lead to LBO. These events were detected in flame optical [21] and
acoustic [22] emissions using simple non-intrusive sensors. The precursor events were
observed to occur more often as the combustor’s stability margin (Φ-ΦLBO or Φ/ΦLBO)
was reduced. Hence, the average occurrence rate of precursor events was shown to
provide a measure of the combustor’s proximity to blowout. Active control of LBO was
also demonstrated by actuating a pilot fuel responding to number of events occurring in a
unit time window.
Spectral methods based on estimation of relative spectral power at low
frequencies, e.g. 5-50 Hz, was also demonstrated to provide a measure of LBO margin by
Nair and Lieuwen [22] and Prakash et al. [23], in the same gas fueled combustor
employed by Thiruchengode et al. [21]. Precursor events having relatively long durations
(20-50 ms) and with low occurrence rates (2-5 sec-1) contribute to increased power at low
frequencies. A similar approach, low frequency tone increase in the acoustic signature
from the combustor of a ground based gas turbine engine, was used to estimate the
probability of incipient blowout of the engine by Taware et al. [24].
Both event based and spectral methods were shown to be capable of approaching
LBO detection in a non-premixed liquid-fueled combustor operating at atmospheric
pressure [25, 26]. In addition standard deviation of the optical signal was also shown to
indicate LBO proximity in a liquid-fueled combustor by Yi and Gutmark [27]. An
alternate sensing technique to optical and acoustic emissions was explored by Thornton et
al. [28]. They used detection of changes in the combustion products’ electrical properties
to detect LBO precursor events. They monitored the current through low voltage
electrodes integrated into the fuel nozzle. During precursor events, there is presumably a
decrease in the ionization level, producing a decrease in the current. Li et al. [29]
employed a tunable, diode laser absorption sensor to detect LBO margin in a lean,
partially premixed model turbine combustor. They monitored temperature fluctuations
10
associated with the localized and temporary extinction events and used a spectral power
approach (0-50 Hz) to analyze the sensor output. For the most part, these techniques rely
on the increased unsteadiness of the heat release associated with the occurrence of
precursor events.
Among all the possible methods for LBO margin sensing, the event based method
is preferable as it has the best time response. Statistical and spectral methods require
signal data over sufficient time for reliable LBO margin sensing. On the other hand,
events can be detected as soon as they occur and better time response can be achieved.
In addition to LBO (static instability), lean combustion often gives rise to
thermoacoustic combustion instabilities (dynamic instability). Dynamic instabilities are
associated with high amplitude pressure oscillations resulting from closed-loop coupling
between pressure oscillations and heat release oscillations. Lean operation with a high
degree of premixing creates a higher risk of dynamic instability problems. For example,
combustors employing lean combustion are designed to operate with mostly primary air.
This reduces the area of the dilution holes that help damp acoustics [30]. In addition, the
combustion process has increased sensitivity to perturbations in equivalence ratio at lean
conditions near blow-out [31]. In a premixed system operating very close to blowout,
equivalence ratio oscillations have been shown to cause periodic flame extinction giving
rise to low frequency oscillations with high amplitudes [30]. Though combustion
dynamics are suppressed to acceptable levels by careful design of practical combustors,
moderate to low levels of dynamics may continue to exist.
The high amplitude dynamic instabilities that can exist near LBO can also trigger
LBO. For example, Snyder and Rosfjord [32] observed that decreasing fuel flow rate
increased instability amplitudes and subsequently caused blow off in a turbine engine. In
stationary gas turbines, low frequency dynamics (0-100 Hz) often referred to as chug or
LBO tone are observed to cause LBO [10, 33]. Similarly in a backward facing step
11
combustor, Cohen and Anderson [34] observed that reducing the equivalence ratio can
produce low frequency instability, with amplitudes rising gradually until blowout
occurred. Besides the dynamic instability itself, control methods used to suppress
dynamics, such as fuel flow rate modulation and fuel spatial distribution control, could
also lead to blow out.
1.3. Overview of Present Work
This thesis is motivated by the goal of improving the robustness of LBO margin
sensing by investigating it under a wider range of scenarios. LBO margin sensing studies,
until now, were conducted under dynamically stable operating conditions. However as
lean combustion often has pronounced dynamic instabilities, even near blow out limits,
LBO margin sensing in the presence of high amplitude dynamic instability needs to be
investigated. In addition, most of the previous LBO margin sensing studies have been
performed at atmospheric pressure under non-realistic engine operating conditions,
primarily in premixed, gas-fueled combustors. Hence, LBO margin sensing needs to be
studied under more realistic engine operating conditions, i.e., at elevated pressure and
preheat operation. The thesis addresses these issues by investigated LBO margin sensing
in more complex scenarios, i.e., under dynamically unstable conditions and elevated
pressure and preheat operation. LBO margin sensing under dynamically unstable
conditions is investigated in a in a premixed, swirl-stabilized combustor similar in design
to ground power gas turbine combustors, operating at atmospheric pressure. LBO margin
sensing at elevated pressure and preheat temperature is examined in a liquid-fueled low
NOx design combustor similar to future aircraft engine combustors.
Though in previous studies, LBO margin sensing and control was demonstrated
during slow transients in combustor operating conditions, its limitations during fast
transients have not been investigated. Since precursors are discrete events occurring at a
small non-constant rate (typically 1-10 per second in the previous studies), it is not clear
12
whether any event would occur before LBO in a sufficiently fast transient. Therefore, the
thesis develops a methodology to estimate the limits on transient rates such that precursor
events occur before LBO. As mentioned earlier, employing LBO margin sensors can
potentially improve transient response of engines during rapid decelerations. To this end,
the thesis examines the transient response improvements and tradeoffs in implementing
precursor event based LBO margin sensing in a model turbofan engine during rapid
deceleration transients.
Chapter 2 of the thesis gives a more detailed background on swirl combustion,
lean blowout, previous LBO margin sensing studies and conventional LBO prevention in
turbine engines. Chapter 3 describes the experimental setups and modeling approaches
used in the present study. In Chapter 4, LBO margin sensing under dynamically unstable
conditions and at elevated pressure and temperature operation using optical
(chemiluminescence) signals is investigated. Chapter 5 investigates the issues in using
acoustic signals for LBO margin sensing. In Chapter 6, new approaches are presented for
examining LBO margin sensing during rapid transients, and the results of one such
analysis are reported. Chapter 7 presents conclusions and contributions of the current
work along with suggestions for future investigations.
13
CHAPTER 2
BACKGROUND AND PREVIOUS WORK
This chapter provides a review of issues related to lean blowout and its margin
sensing in swirl-stabilized combustors. The first section describes flow field and flame
configurations in swirl combustors. The second section discusses the physical
mechanisms associated with lean blowout. The third section describes previous work in
LBO margin sensing. The fourth section discusses combustion dynamics and the fifth
section covers issues related to control of aero engines.
2.1. Swirl Stabilized Combustion
Practical combustion systems are usually designed to produce high heat release
rates in compact volumes, i.e., high power densities, in order to reduce size, weight and
cost. In addition, combustors need to be stable over a wide range of combustor loadings
and operating conditions. Moreover, high combustion efficiency with low emissions is
required. Designers are faced with the challenge of optimizing a combustor to meet all
these objectives.
To achieve high power densities, practical combustors often employ high
pressures along with high reactant velocities. For gas turbines, the high pressures are also
necessary to increase cycle efficiencies; typical gas turbine combustors operate at
pressures up to 30-40 bar and employ average flow velocities in the range 10-35 m/s [7].
Flame stabilization in high velocity flow can be challenging. Generally, a premixed flame
can be stabilized if flame speed can be matched to flow speed at some point in the flow.
However laminar flame speeds, having a range of about 10-100 cm/sec [35], are much
lower than typical flow velocities. Therefore, additional stabilization approaches are
required to ensure flame stability.
14
For high velocity flows, flow recirculation is usually employed for flame
stabilization. Recirculation regions hold hot products and radicals and promote mixing of
reactants with them. Thus recirculation provides a continuous supply of heat and radicals
for ignition of reactants and increased flame speed. Combined, these effects lead to
enhanced flame stabilization. Commonly used recirculation configurations are: 1) flow
recirculation created by vortex break down of a swirling flow; 2) the wake of a bluff
body; and 3) sudden expansion created by a step increase in flow area. Other novel
methods that do not rely on recirculation are also being investigated for flame
stabilization, for example the low swirl burner [36]. It uses low velocity regions created
by divergence of flow with a low amount of swirl for flame stabilization.
Gas turbine main combustors predominantly employ swirl-stabilization because
swirl: 1) improves flame stability; 2) achieves fast mixing between fuel and air; 3)
produces high entrainment of ambient combustor fluid [37]; and 4) results in high power
densities. Swirling flows produce high levels of turbulence [8, 38], which increase
turbulent burning velocities producing high heat release in a small volume. In addition,
the high turbulence levels help faster fuel-air mixing, reducing emissions. The tangential
velocity component of the swirl increases aerodynamic forces on fuel spray resulting in
faster spray breakup and smaller droplets. The recirculation zone of the swirling flow
creates low velocity regions, and stagnation points where flame speed can match flow
speed, for improved stabilization [39].
Swirl is created by imparting a tangential velocity component to the flow using
swirl vanes, axial plus tangential flow entry or pure tangential flow entry in to a chamber
[8]. The strength of a swirling flow is characterized by its swirl number (S), defined as
the ratio of axial flux of tangential momentum to the axial flux of linear momentum. For
sufficiently strong swirl (S>0.6), flow reversal in the vicinity of the central axis is created
forming an inner recirculation zone (IRZ). In combustors, formation of an IRZ is usually
15
triggered by employing a sudden expansion geometry. Formation of the IRZ occurs
through the following physical processes [40]. Swirling flow creates a radial pressure
gradient due to centrifugal forces having low pressure near the axis. As the flow expands
through the sudden expansion geometry, axial decay of tangential velocity and the radial
pressure gradient occur. The process creates an adverse axial pressure gradient around the
axis, which in turn causes flow reversal. Formation of a recirculation zone is commonly
referred to as vortex breakdown. Vortex break down is defined as an abrupt change in the
structure of the swirling vortex core resulting in stagnation points and flow recirculation
[41]. Vortex breakdown can take several forms depending on swirl strength and
Reynolds number. For example, bubble, spiral and helical modes of vortex breakdown
have been observed [42, 43]. The type of vortex breakdown characterizes the
recirculation zone and evolution of the flow field further downstream. The bubble type of
vortex break down is the most common form observed in combustors.
An illustrative swirling flow field having bubble type vortex breakdown is shown
in Figure 4 for a typical combustor configuration commonly employed in low NOx
premixed combustion systems. The combustor has an annular swirling flow, around a
cylindrical center body, issuing into a sudden expansion [7, 44]. The flow field consists
of an inner recirculation zone (IRZ) created by bubble type vortex breakdown. In
addition, it consists of an outer recirculation zone (ORZ) over the expansion region (e.g,
a backward facing step). Along the boundaries between recirculation zones and the
swirling jet, sharp gradients in axial and tangential velocity exist, creating regions of high
flow shear. These high shear regions are called the inner and outer shear layers (see
Figure 4). The IRZ may be closed, i.e., backward flow only over a short distance, or the
backward flow may extend throughout the length of the combustor tube, around the
central axis. Employing a converging geometry for the flow exit affects the extent of
backward flow region, and suppress it [45]. Besides the IRZ structure shown in Figure 4
with reverse flow in the entire IRZ region, in some cases IRZ may have a two cell
16
structure with forward velocity in the vicinity of the central axis [42]. In addition to the
IRZ created by vortex breakdown, a recirculation zone is formed in the wake of the
center body. The center body recirculation may merge with the IRZ forming a continuous
reverse flow region. The flow fields seen for isothermal (and isodensity) flow fields can
be significantly altered by the heat release associated with combustion, mainly due to
dilatation effects. Combustion decreases effective swirl strength, i.e., swirl number, by
increasing axial velocity. It reduces the strength of the recirculation zone by lowering the
amount of the recirculating fluid [8]. In addition, combustion may suppress the existence
of backward flow over a long distance [45] or it may promote the formation of a two cell
IRZ.
The flow structure discussed above is based on time average characteristics. Due
to high shear in axial and azimuthal shear layers, large scale spatial and temporal
fluctuations in the flow field occur [8]. In the shear layers, large scale coherent structures
are formed, due to Kelvin-Helmholtz (K-H) instabilities, resulting from the combined
effect of axial and azimuthal shear layers [46]. The coherent structures consist of
concentrated vortex rings that convect downstream in a helical fashion. The coherent
structures modify the flow field and combustion process. For example, the coherent
structures have been observed to induce large asymmetry in the flow about the axis [47].
In addition, they can modify the combustion process by wrapping the flame around them,
resulting in an increase in unsteady burning. As they travel downstream, the coherent
structures break down into smaller, less organized turbulent structures. Besides the
coherent structures, swirl flows are often observed to produce precessing vortex cores
(PVCs). PVCs are associated with the CRZ precessing around the geometrical axis of the
combustor. They are helical in nature, with low pressure in the core, and are wrapped
around the CRZ. The occurrence of PVCs is mainly dependent on swirl number and
combustor geometry. In addition, the heat release, the type of combustion, i.e., premixed
or non-premixed, and the degree of flow confinement influence PVCs. PVCs strongly
17
affect the flow field, for example they displace the swirling vortex core off the axis and
result in non-uniform azimuthal velocities with higher velocities near the combustor wall.
They typically exist for a downstream distance of 1-2 combustor diameters, starting from
the inlet, before breaking up. In addition to the organized structures, the flow includes a
high degree of random turbulent velocity fluctuations, with peaks in the inner and outer
shear layers.
Inner Recirculation
Zone (IRZ)
Inner Shear Layer

Recirculation zone
(ISL) boundaries
Outer Recirculation
Outer Shear Zone (ORZ)
Layer (OSL)
Figure 4. Illustration of a swirling flow field structure.
Depending on the combustor geometry, reactant mixture properties and operating
conditions, different flame configurations exist in swirl-stabilized combustors. The
configurations often have a pronounced effect on flame static and dynamic instability,
emissions and wall heat transfer. Example flame configurations in premixed operation,
for the combustor geometry in Figure 4, are shown in Figure 5. In configuration (a), a
flame is present in inner and outer shear layers. The flame separates cold reactants in the
swirling jet from hot products in recirculation zones. The flame is attached to the center
body due to stabilization by the recirculation zone of the center body or by the shear
18
layer. In configuration (b), there is no flame in the outer shear layer, while there is an
inner shear layer flame similar to that in configuration (a), though longer in length to burn
all the fuel. For sufficiently lean mixtures, configurations (c) and (d) have been
observed. Shifts between flame configurations during combustor operation may occur
due to changes in equivalence ratio, fuel composition, preheat temperature and other
conditions. The primary cause of configuration shift is due to variation of reactant
mixture flame speed or extinction strain rate with changing conditions. The shifting
process is often observed to be sudden [48] and sometimes with oscillations [49]. For
example, the flame can shift from configuration (a) to (b) when equivalence ratio and
preheat temperature are reduced. The configuration can shift back for opposite changes in
the conditions. By reducing equivalence ratio, flames have also been observed to shift
from configuration (a) to (d), progressively [49, 50]. These flame configurations are on a
time-average basis. Instantaneous flame shapes will have severe corrugations due to
turbulent, vortical structures wrapping the flame around them [51, 52]. With non-
premixed combustion, flame configurations are more complex, influenced by fuel-air
mixing. Generally, the flame surface may envelope the inner recirculation zone.
(a) (b) (c) (d)
Figure 5. Illustrative flame configurations in premixed swirl combustors.
Shifts in flame configurations can be analyzed from the perspective of strain rates
experienced by flames relative to extinction strain rates of the reactant mixture. Flame
19
straining is essentially caused by species and thermal diffusion non-aligning with flow
stream lines and resulting in enthalpy and stoichiometry modification, locally. Such
conditions occur due to flame curvature or flow shear upstream of a flame [53]. As an
example, a flame can develop local curvature due to vortical turbulent structures whereas
velocity gradients cause flame aerodynamic shear. Flames can withstand strain only up to
a certain level, denoted the extinction strain rate. Extinction strain rate tends to decreases
with reduction in equivalence ratio below stoichiometric conditions. Therefore lean
conditions conducive to low NOx operation produce flames that are more susceptible to
extinction. Extinction strain rate of a lean, laminar, methane air flame varies from 500 to
100 s-1 between equivalence ratios of 0.6 to 0.5 [53] . The swirl flames described above
experience high strain rates in the inner and outer shear layers, sufficient to cause
extinction of lean flames. For example Wicksell et al. [51] and Stohr et al. [54] observed
strain rates of the order of 1,000 s-1 in shear layers. Flame extinction in stabilizing
regions due to high strain rate would result in flame configurations shift. The flame may
find new stabilizing locations where it experiences low strain rates.
2.2. Lean Blowout
The operating conditions that influence flame blowout in a combustor include
fuel-air ratio, pressure, preheat temperature, velocity and fuel composition. In addition,
blowout limits vary greatly with combustor design. Identification and understanding of
the key physical processes responsible for LBO can help in developing better margin
sensing approaches.
The amount of heat released from chemical reactions inside a combustor is a
function of the residence time of the flow inside the combustor, due to the finite time
required for completion of chemical reactions. The finite chemical rate can be
characterized by a chemical time scale. If the heat released within the flow residence time
inside a combustor fails to increase reactant energy (temperature) above the minimum
20
activation energy required for combustion, it gets extinguished. This approach has been
used for predicting blowout conditions using well stirred reactor (WSR) models [55]. The
two important time scales of the problem are the flow residence time and the chemical
time. Dahmkohler number, defined as ratio of flow residence time to chemical time, can
typically capture blowout trends. The applicability of well stirred reactor models for swirl
combustion requires that reactants are mixed with products, such as combustion in the
distributed reaction zone regime [56] of turbulent combustion. It has been speculated that
high turbulence levels in the inner recirculation zone and shear layers of a swirl-stabilized
combustor result in nearly perfect mixing of reactants and products. Few studies report
successful prediction of blowout conditions using well stirred reactor models [14, 57]. In
addition, Domkohler number scaling has been reported to be useful in blowout
predictions [58, 59]. However, a study by Sloan and Sturgess [60] using well stirred
reactors at small scales in a complete simulation, failed to predict flame liftoff and
blowout conditions in a non-premixed swirl combustor.
Flame blowout may also occur due to insufficient time for ignition of reactants in
the shear layer between a recirculation zone and reactant stream [16, 61, 62].It has been
suggested that hot gases in the recirculation zone mix with reactants in the shear layer,
auto igniting them, and thus provide a flame stabilization point [63]. If the ignition delay
time is large compared to the contact time of reactants with hot gases, blowout will occur.
The physics can be scaled using a Damkohler number defined as the ratio of flow
residence time to the ignition delay time, for blowout prediction. This methodology has
been mainly applied in bluff body stabilized flames.
A third alternative mechanism considers extinction of flames in a turbulent flow.
The basis for the approach comes from recent observations that practical combustors
typically operate in flamelet or thin reaction zone regimes rather than in the distributed
reaction zone regime [64]. Studies employing planar laser induced fluorescence of CH
and OH reported clear existence of laminar or thickened flamelets even in highly
21
turbulent flows [65]. Thickened flamelets are caused by small scale eddies (for example
eddies on the Kolmogorov scale), penetrating the preheat zone and transporting cold
reactants to it. If the turbulence is high enough, the smallest eddy scales can penetrate the
reaction zone and mix products with reactants, creating distributed reaction zones.
However, evidence for the existence of distributed reaction zones is rare [64]. Therefore,
well stirred reactor models which assume mixing of products with reactants may not
describe the controlling physics. An appropriate physical description for blowout should
consider extinction of flamelets and subsequent alterations in the flow field leading to
blowout.
In a turbulent combustion, local flame extinction occurs due to two processes. The
first is straining of flamelets by flame-vortex interactions and the second is quenching by
turbulent transport. Medium and large scale eddies in a turbulent flow impose severe
flame strains due to wrinkling and shearing of the flame. The strain may cause local
extinctions if it is imposed for a sufficiently long time. Flames can withstand
instantaneous strain rates, much above the extinction strain rate (obtained from steady
state data) if only imposed for a brief period [64]. Another process of flame quenching by
turbulence is due to smallest scale eddies penetrating the reaction zone and quenching the
flame by transporting cold reactants in to the reaction zone. Such conditions occur if the
turbulence intensity is high or the mixture is lean enough, such that the size of small scale
eddies is about the thickness of the reaction zone. If the smallest scale eddies could only
penetrate the preheat zone, extinction may not occur. By reducing equivalence ratio,
extinction strain rate decreases and flame thickness increases, increasing the likelihood of
strain extinction and turbulence extinction respectively. By increasing mean velocity
turbulence levels usually increase, increasing both strain and turbulence extinction. Of
these two extinction mechanisms, strain induced extinction may dominate, as few studies
reported local extinction of thin reaction zones before broadening by turbulence [66].
22
Prior to blowout, local extinctions can randomly occur in a flame, which may
grow in size or shrink and may be convected by the mean flow. Though there is no direct
interaction between the recirculation zone and the flame during a normal continuous
flame sheet, interactions become important when holes appear in the flame sheet. For
example, the recirculation zone may help in re-ignition of reactants at the hole locations,
restoring the flame. On the other hand, holes allow (cold) reactants into the recirculation
zone, which may weaken it. If the rate of supply of reactants into the recirculation zone
exceeds their rate of conversion to products , cooling of the recirculation occurs. This
would reduce its ability to restore flame along the holes, thereby further increasing the
supply of reactants into the recirculation zone. An ultimate condition would be reached
when the holes cannot be restored and complete extinction of the flame sheet occurs,
leading to blowout.
In a bluff body stabilized flame, Chaudhuri et al. [67] observed that a flame front,
which usually envelopes the shear layer during a stable condition, enters the shear layer
near blow off. The flame in the shear layer is subject to severe straining by shear layer
vortices resulting in local extinctions. Fresh reactants enter the recirculation zone,
forming flame pockets which in turn can re-ignite the flame in shear layers. Such
extinction and re-ignition processes can happen several times, and complete blowout
occurs when the flame pockets fails to re-ignite the shear layer flame. Similar
observations were made by Kariuki et al. [68] in a bluff body stabilized flame, where
they observed entrainment of fresh reactants into the recirculation zone from
downstream, severe local extinctions in the flame front and formation of flame pockets
inside the recirculation zone. They observed that total flame blowout occurred when the
flame at the bluff body edges was lost.
23
2.3. LBO Margin Sensing: Previous Work
Local flame extinctions and re-ignitions occur in lean conditions before blowout
[69]. However the local extinctions do not lead to immediate and complete flame
blowout. Local extinctions can result in large scale unsteadiness in the flame [15, 70].
The extinctions and the associated unsteadiness have been used for detecting the
proximity to LBO.
Based on flame partial extinctions, i.e., local extinctions, followed by re-ignition
prior to LBO, Thiruchengode [69] sensed approaching LBO in a premixed swirl
combustor. He observed intermittent large scale partial extinction of flame in the inner
shear layer and its subsequent re-ignition before LBO. The partial extinction and re-
ignition resulted in large scale unsteadiness in the flame configuration. A stable flame
during nominal combustor operation and a temporary extinction and re-ignition process,
observed in high speed flame images are shown in Figure 6. Near LBO, combustor
operation is characterized by a stable flame with intermittent extinction and re-ignition
events. During an extinction and re-ignition event, there is a large scale flame loss at the
bottom of the combustor and most of the fuel is burnt downstream of this region.
Frequent occurrence of such events preceded complete flame blowout. Therefore these
were called LBO precursor events.
(a) (b)
Figure 6. (a) Stable flame during nominal operation (b) Sequence of flame images
during an extinction and re-ignition event; image separation 4 ms. (Ref [69])
24
The precursor events were detected non-intrusively by monitoring flame OH*
chemiluminescence emissions. Sample precursor events in the OH* optical signal time
trace are shown in Figure 7. In the figure, signal features with a conspicuous drop in
signal amplitude are precursor events, as extinctions result in a decrease in heat release
and the amount of light emitted by species created through chemical reactions. In this
previous work, events were detected using a double thresholding approach. An event is
declared to start when the signal drops below a first (lower) threshold, and is declared to
end when the signal rises above a second (higher) threshold. With a single threshold,
inherent noise in the signal can cause the signal to cross the threshold multiple times,
causing a single event to be counted as multiple (shorter) events. A minimum event
duration constraint was also imposed for more robust event detection. The threshold
levels are based on recent signal outputs, i.e., a threshold is given by -, where  is the
time-localized signal mean,  is the time-localized signal RMS and  is a constant.
Setting the threshold relative to the signal mean makes the event sensing approach less
sensitive to “slowly” varying conditions such as engine power and sensor drifts. Making
the threshold a multiple of the signal RMS also reduces false event detection that could
be caused by combustion or detector noise.
Upper threshold
Amplitude (a.u.)
Time (ms) Lower threshold

Figure 7. Precursor events in the OH* optical signal along with thresholds. (Ref.
[69])
Results of event identification, with the double thresholding method, time
averaged, are shown in Figure 8. The number of events occurring in a given time period,
25
on average, increase as the combustor gets closer to its LBO limit. Therefore event
occurrence rate can be used as a measure of a combustor’s proximity to blowout.
1.8
Average number of events/sec
1.5
1.2
LBO limit
0.9
0.6
0.3
0
0.74 0.75 0.76 0.77 0.78 0.79 0.80 0.81 0.82

Equivalence ratio (Φ)
Figure 8. Variation of average number of precursor events with equivalence ratio.
(Ref. [69])
In addition to using flame chemiluminescence, acoustic radiation generated by the
flame has also been used for LBO margin sensing [71]. Acoustic emission from the
combustion process is proportional to the time derivative of heat release rate. During
extinction and re-ignition events, the sudden drop and then rise in heat release produces a
corresponding acoustic signal. A representative precursor event in the acoustic signal is
shown in Figure 9. The time scale of this precursor event feature is ~5 ms. As with
optical sensing, acoustic detection allows for non-intrusive sensing with a simple sensor,
which is advantageous for practical considerations. Event occurrence rate can be
determined either by simple thresholding or by first applying filtering techniques such as
a wavelet transformation of the signal using a first order Gaussian wavelet, as this has a
similar shape to the precursor event [71].
26
Amplitude (a.u.)
Time (s)
Figure 9. Precursor events in the acoustic signal. (Ref. [71])
Power Spectral Density (a.u.)
Frequency (Hz )
Figure 10. Power spectra of acoustic signals in a premixed swirl combustor. (Ref.
[71])
In addition to event identification based methods, spectral methods have also been
demonstrated to be capable of LBO margin sensing [23, 69, 71]. The relative spectral
power in the low frequency range, e.g., 0-50 Hz, of the optical and acoustic signals was
shown to increase near LBO. Precursor events usually have long durations and low
occurrence rates. These two factors contribute to a relative increase in spectral power at
low frequencies. Thus monitoring of the fractional power in the low frequency range has
been suggested as a LBO proximity parameter [71].
27
Events/s
Φ Pilot valve
Time (s)
Figure 11. Response of an LBO control system to varying operating conditions. (Ref.
[69])
An example of LBO control using precursor event detection is shown in Figure 11
[69] . The control system consists of a pilot fuel valve and responds to precursor event
count. Here the combustor equivalence ratio was reduced slowly below its normal LBO
limit. However, the event count increased as the combustor got closer to the LBO and the
control lsystem opened the pilot fuel valve to stabilize the combustor.
2.4. Combustion Dynamics
Combustor operation in lean conditions, including near the LBO limit, can
sometimes include significant levels of dynamic instability. Self-sustained oscillations in
pressure and velocity near natural acoustic frequencies of the combustion chamber,
produced by closed loop coupling between unsteady heat release and pressure
fluctuations, are referred to as dynamic instability. In lean premixed gas turbine
combustors, coupling between pressure oscillations and heat release oscillations takes
place mainly by flame-acoustic interactions or acoustic-fuel feed system coupling [72].
Flame and acoustic wave interactions involve perturbations in flame area, position or
28
stretch, resulting in heat release perturbations, which provide feedback to the acoustics.
Instability mechanisms involving the fuel feed system result from modulation in fuel flow
rate caused by pressure oscillations in the fuel injection/premixing section. Fuel flow rate
oscillations result in equivalence ratio oscillations, which get convected to the flame in
the combustor resulting in heat release oscillations. It should be noted that an instability
mechanism involving equivalence ratio oscillations could simultaneously exhibit flame-
acoustic interactions.
2.5. Passive Control of LBO in Aircraft Engines
Current aircraft engines employ a passive control method for avoiding LBO, as
the actual LBO limit cannot be determined. The method essentially takes a measure of
combustor equivalence ratio and compares it with a predetermined LBO limit. For
engines in service, the fuel flow rate can be metered accurately. However, the airflow rate
through the combustor is not measured directly [73]. The combustor airflow rate can be
estimated using compressible flow properties of the turbine inlet nozzle guide vanes.
Flow through the inlet guide vanes is nearly choked over most of the operating range of
an engine [74]. Therefore, the mass flow rate of air through the combustor is proportional
to P/√T where P and T are the combustor exit static pressure and temperature. Typically,
combustor exit static pressure is not measured due to high temperatures. Nevertheless,
compressor exit static pressure can be approximated by the combustor inlet static
pressure, due to the relatively small pressure drop within the combustor. While the
combustor exit temperature can be monitored, the measurement has a slow time response
due to heat shielding of thermocouples [75]. However, due to relatively smaller changes
in √T compared to P over the operating range of an engine, airflow rate is approximately
proportional to compressor exit static pressure. Thus, the ratio of fuel flow rate to the
compressor exit static pressure (often denoted as the Ratio Unit, RU) provides a measure
29
of fuel-air ratio (see Appendix A for the derivation). Engine control systems commonly
employ a minimum limit (RU limiter) on this ratio to prevent LBO [75, 76].
The primary function of the aircraft engine control systems is to achieve a desired
thrust while keeping the engine with in its safe limits. Since thrust cannot be measured
directly, shaft speed is usually taken as a measure of the thrust. The control system varies
fuel flow rate (control variable) to achieve the desired shaft speed (controlled variable).
The control system architecture consists of a set point controller, to achieve the desired
shaft speed, and several protection logic controllers to keep the engine within safety
limits. Set point controllers commonly use proportional-integral (PI) control logic, which
takes an error in the shaft speed as an input and outputs fuel flow rate. Instead of fuel
flow rate, the control systems usually use RU as the control variable due to its better
control performance [77]. RU multiplied by the current compressor exit static pressure is
commanded as the fuel flow rate for the next time step.
As described in Chapter 1, combustors get closer to LBO limit during power
reduction deceleration transients as the fuel flow is reduced without a proportional
reduction in the air flow. An engine control scheme during a deceleration transient, along
with the acceleration transient, is illustrated in Figure 12. Engines go through
deceleration transients when the throttle setting is reduced, demanding a lower speed than
the current speed. The negative speed error results in the set point controller commanding
a lower RU and moving the engine along the path 1-2. At point 2, the minimum RU limit
is reached, and a MAX select strategy chooses the maximum of the fuel flow rates given
by the set point controller and the minimum RU limiter. The engine moves along 2-3,
until it reaches a point where the set point controller commands a higher fuel flow rate
than the fuel flow rate corresponding to the minimum RU limiter. At this point the
control switches back to the set point controller and the final desired speed is achieved.
30
Surge limit
Max RU line
1
W f /Ps3 (RU)
3 2
Min RU line
Idle Full speed

N/Nmax (%)
Figure 12. Illustrative engine control schedules during transients. (Ref. [75])
31
CHAPTER 3
EXPERIMENTAL AND MODELING APPROACHES
This chapter describes the equipment and approaches used to conduct the
experiments described in this thesis, as well as the modeling methods employed in the
current study. LBO margin sensing is examined in two different types of combustors. The
first is a gas-fueled (natural gas) combustor, emulating a single burner in a ground-based
gas turbine combustor. The second is a liquid-fueled Lean Direct Injection (LDI)
combustor more representative of next-generation low NOx aeroengine combustors. The
first section describes the gas-fueled combustor experiment. The second section describes
the liquid-fueled combustor setup. The third section describes the model used to simulate
a turbofan engine and control system for exploring the impact of LBO margin sensing on
deceleration transient time response.
3.1. Gas-Fueled Combustor Setup
3.1.1. Combustor Design and Flow Facility
In order to study LBO margin sensing in the presence of combustion dynamics, a
gas fueled combustor is employed. The combustor is swirl and dump stabilized and
operates at atmospheric pressure. A schematic diagram of the combustor setup is shown
in Figure 13. Air enters the combustor through a choked valve in order to isolate the air
supply from combustor disturbances, specifically pressure oscillations. The air passes
through a 22 mm diameter tube, having two axial swirlers with straight blades. The first
swirler has a vane angle of 35, while the second has a higher vane angle of 50. Thus the
theoretical swirl number of the inlet flow is 0.842 [8]. The swirlers are placed 80 mm
apart with the second swirler 25.4 mm from the combustor dump plane. Each swirler has
an outer diameter of 22.8 mm and a hub diameter of 6.3 mm. The inlet also incorporates a
10 mm diameter (cylindrical) centerbody having a length of 25 mm for enhanced flame
32
stabilization. The combustor liner is formed by a quartz tube, which can sustain high wall
temperatures and permits transmission of ultraviolet (UV) radiation. It has an inner
diameter of 70 mm and outer diameter of 75 mm with a length of 600 mm. The quartz
tube sits inside a groove made around the combustor inlet, permitting optical access to
the entire flame.
Microphone
70mm
600mm
Optical Fiber
Swirler
Fuel Fuel Plenum
Swirler
Fuel Injection
Orifices Choked Valve
Fuel
Air
Figure 13. Schematic diagram of the gas fueled combustor setup.
This combustor was used as it can exhibit dynamic instability in lean conditions
near blowout. The combustor is configured to have combustion instabilities of two kinds:
with and without equivalence ratio oscillations. This is achieved through two types of
fuel injection. Gaseous fuel can be injected into the air stream either far upstream before
the choked orifice, or after the choked orifice just upstream of the second swirler. By
injecting fuel before the choked orifice, pressure perturbations cannot excite fuel-air ratio
oscillations. In addition, the large distance between the combustor and the injection
location produces well-mixed reactants. In the second method, fuel enters the air stream
33
between the two swirlers, through multiple injection orifices, none of which are choked.
The orifice module has three rows of orifices along the axis with 12 orifices in each row
spaced symmetrically around the axis. This arrangement gives a total number of 36
orifices (see Figure 14). The distance between orifice rows is 3.8 mm and each orifice has
a diameter of 2.5 mm. The closest row of orifices to the combustor inlet is located 68 mm
upstream of the dump plane. The large number of injection orifices is intended to achieve
substantial premixing, while still allowing for feedback from combustor oscillations to
perturb the fuel flow rate and therefore the incoming equivalence ratio.
2.5mm
3.8mm
22mm
Figure 14. Fuel injection orifice module.
Combustion air flow is delivered from the building air supply line at 250 psig,
with a pressure regulator used to ensure a constant supply pressure. Similarly, natural gas
comes from the building supply at 125 psig and then passes through a pressure regulator.
In cases where methane is used, it is supplied from compressed gas bottles connected to
the combustor’s standard fuel supply line. A schematic diagram of the flow supply,
control and monitoring system is shown in Figure 15. Air and fuel flow rates are
measured with rotameters and with pressure gauges monitoring line pressures. For fine
control of fuel flow rate during combustor operation near LBO, a rotameter with a
resolution of 0.24 SLPM (0.5 SCFH) is used in parallel with the main fuel rotameter,
34
which has a resolution that is ten times coarser. For the nominal airflow rates used in the
experiments, the equivalence ratio resolution is approximately 0.003. The fuel supply line
is split into two paths with a three-way valve to rapidly switch between far upstream
injection and close injection modes without shutting down the combustor. The nominal
average cold flow axial velocity in the combustor is 4.5 m/s, and the nominal combustor
power is 32 kW (0.11 MBTU/hr). Based on complete combustion and no heat losses, the
bulk average exit axial velocity of the product gases would be 23 m/s.
To combustor
Pressure Gauge
Rotameter
Air
Far upstream injection

Pressure Gauge
Pressure Regulator Valve
Close injection
Fuel
Rotameter
Figure 15. Schematic diagram of fuel and air supply, control and monitoring setups.
The combustor exhibits three distinct flame configurations based on equivalence
ratio and flow velocity. The three configurations for different equivalence ratios at a fixed
air flow rate (corresponding to a cold flow combustor velocity of 2.5 m/s) are shown in
Figure 16. For these images, the combustor was made dynamically stable by reducing
the combustor tube length to 300 mm from the standard (600 mm) value, and by
employing the far upstream fuel injection mode. However, even during dynamically
unstable conditions, the basic flame configurations are essentially the same.
35
(a) (b) (c)
Figure 16. Flame configurations in the combustor for different equivalence ratios at
a cold flow velocity of 2.5 m/s: (a) Φ=0.88, (b) Φ=0.79, (c) Φ=0.66.
For equivalence ratios above 0.88, the flame exists in both the inner and outer
shear layers as shown in Figure 16(a). In this configuration, the flame is attached (or sits
very close) to the centerbody. When the equivalence ratio is reduced slightly below 0.88,
the flame in the outer shear layer intermittently extinguishes at a low frequency. When
the equivalence ratio is further reduced to 0.79, the outer shear layer flame completely
disappears leaving only the inner shear layer flame as seen in Figure 16(b). In this
configuration, the flame is longer. By decreasing equivalence ratio further, the flame
shifts to the configuration shown in Figure 16(c), where it is quite long and extends
36
beyond the exit of the combustor. With a sufficiently short tube, e.g., 150 mm, this flame
is not stable and blows out. With increasing flow velocity, the equivalence ratios where
the configuration transitions occur shift to higher values. Similar flame configurations
have been observed in other studies using similar combustor geometries. For example,
Muruganandam [69], Bradley et al. [49], and Chterev et al. [50] reported similar flame
configurations.
3.1.2. LBO Sensing and Diagnostics
Optical and acoustic radiation signals are acquired from the combustor along with
simultaneous high speed chemiluminescence images. A fused silica optical fiber, with a
diameter of 365 µm and a cone angle of 24 was used for collection of optical emissions
from the combustor. With the fiber located 60 mm from the center of the combustor, the
optical detection region extends across the width of the combustor and 23 mm in the axial
direction at the centerline (see Figure 13 ). The optical radiation first passes through an
interference filter centered at 308 nm with a full-width-half-maximum (FWHM) of
10 nm, which corresponds to the primary region of OH* emission. A miniature metal
package photomultiplier (PMT, Hamamatsu H5784-04), with a built-in amplifier, is used
to detect the optical signal.
Acoustic radiation from the flame is monitored with a condenser microphone
(Bruel and Kjaer type 4939, flat frequency response up to 40 kHz) located 0.30 m above
and 0.5 m radially offset from the center of the combustor exit. The output of the PMT
and microphone are captured with a National Instruments A/D conversion board using
the LABVIEW program.
High-speed flame imaging is also employed to elucidate the flame behavior. The
camera is an intensified, high-speed CMOS camera (Videoscope International
Ultracam3), capable of taking up to 20,000 frames/sec with variable intensifier gating.
The intensifier of the camera has a spectral response from ~200 to 700 nm and hence
captures all the visible light from combustion.
37
3.2. Liquid-Fueled Combustor Setup
3.2.1. Combustor Design and Flow Facility
LBO margin sensing studies for combustor operation at elevated pressure and
inlet air temperature are carried out in a liquid-fueled Lean Direct Injection (LDI)
combustor. Regarding the appropriate operating conditions, the question of what can be
considered realistic operating conditions arises. For example, one can consider ground
level (take off) or cruise conditions. A motivating aspect of the current work is low NOx
engines for future supersonic passenger aircraft, which would operate at high altitudes
(stratosphere) around 18.2 km (60,000 ft). Even with very high compressor pressure
ratios, e.g., 40, combustor inlet static pressures would be between 2-5 atm. Hence in the
present work, an LDI combustor operating nominally between 2-5 atm and with an inlet
air temperature of ~700 K (800 F) is used. In addition, LBO is usually a problem at part
power operation or near idle where compressor pressure ratios are low. Therefore 2-5 atm
is close to the combustor pressure for part power operation at cruise altitudes or idle
operation at ground level, for conventional commercial passenger aircraft.
The LDI injector used in the experiments is a single element configuration of a 9-
element LDI injector, developed by NASA Glenn Research Center [5]. The 9-element
LDI injector is shown in Figure 17. The 9-element injector has nine air swirlers with a
fuel injector in the middle of each swirler in a 76.276.2 mm2 overall area. In the present
injector configuration, only the center element is present. To partially match the
performance of a 9-element injector, the surrounding injectors are replaced by co-flowing
air, using a perforated plate.
38
25.4 mm
76.2mm
76.2mm
Figure 17. NASA 9 element LDI injector. [5]
Fuel
nozzle Venturi
Swirler
(a) (b)
Figure 18. (a) LDI element assembly with venturi, swirler and fuel nozzle. (b)
swirler.
An LDI element assembly consisting of a swirler, fuel nozzle and a converging
diverging venturi section is shown in Figure 18(a). The swirler has helical axial blades,
as shown in Figure 18(b), with a vane angle of 60°. It has an outer diameter of 22 mm
and hub diameter of 9.4 mm. Its theoretical swirl number is 1.02 [8]. The swirler is
provided with tip protrusions so that it can be mounted inside the venturi firmly, at a
known and fixed angular orientation. A cross sectional view of the venturi, with
dimensions in millimeters is shown in Figure 19(a). The exit diameter of the venturi is
39
22 mm. Both converging and diverging sections are inclined at an angle 40° with respect
to the axis. A schematic cross sectional view of the fuel nozzle stem is shown in Figure
19(b). It is a simplex type, pressure-swirl atomizer. The nozzle is nominally located
slightly behind the venturing throat, and includes a filter in order to prevent any foreign
matter from blocking the narrow flow passages, such as the exit orifice. It incorporates an
air gap created by a tube with slightly higher inner diameter than the fuel supply tube, in
order to prevent fuel coking (as the heated air flows around the fuel nozzle). The
minimum required pressure differential across the nozzle is 30 psi in order to produce an
acceptable quality spray. The LDI injector was slightly modified from the version
supplied by NASA, but performance tests indicated the same spray behavior as the
original, unmodified version (see Appendix B for more details).
Swirl
Air gap Filter Nozzle tip
(a) (b)
Figure 19. (a) Cross sectional view of converging-diverging venturi section (all
dimensions in mm) and (b) fuel nozzle stem cross sectional view.
The single element LDI injector configuration uses co-flow created by a
perforated plate around the flow through the LDI element. The co-flow is employed in
order to simulate the effect of surrounding elements, absent in a single element
configuration. The perforated plate, along with the venturi, is shown in Figure 20. It has
2.38 mm diameter holes spaced symmetrically around the axis, with equal distance
40
between any two adjacent holes. There are a total of 114 holes, producing an 88%
geometric area blockage.
Venturi
Figure 20. Perforated plate.
A cross sectional view of the single element LDI injector assembly mounted to a
quartz combustor liner is shown in Figure 21. Airflow flow through the LDI element and
the co-flow are supplied from a common plenum. The airflow split ratio between the two
paths is determined by their effective areas. The fuel nozzle is held in a precise position
with a nozzle holder connected to the injector casing. The fuel line outer tube, used for
creating the air gap, is coated with ceramic lining in order to give additional thermal
protection against coking. A hydrogen pilot flame, ignited by a sparking electrode, is
used to ignite the combustor. The pilot fuel tube and the igniter electrode are mounted on
the injector casing lip as shown in the figure. In order to measure acoustic pressure inside
the combustor, access is provided by a 6.35 mm diameter standoff tube mounted on the
injector lip. A cylindrical quartz tube with 80 mm inner diameter and 85 mm outer
diameter acts as the combustor liner. The quartz liner sits in a groove at the injector lip
and is held in place using a metal ring mounted at its downstream end.
41
Fuel nozzle
holder
Perforated plate
Casing Igniter
Air
80mm
Air
304mm
Fuel nozzle Venturi Quartz combustor

Swirler
liner
Acoustic
signal tube
Figure 21. Cross sectional view of the single element LDI combustor.
The LDI combustor is mounted inside a high pressure test rig that can operate up
to at least 20 atm. A cross-sectional view of the high pressure test rig with the LDI
injector mounted is shown in Figure 22. The pressure vessel takes the high test pressure
with no pressure difference across the quartz combustor liner. The vessel has an inner
diameter of 0.193 m and a length of 0.495 m. It is equipped with quartz optical windows
permitting optical signal acquisition and flame visualization. Cooling air at room
temperature passes between the quartz liner and pressure vessel. A gradually converging
section is attached downstream with an orifice at the end to achieve pressurization. The
test rig is controlled remotely from a control room due to safety requirements.
42
Acoustic sensor
Pressure vessel
standoff tube
Exhaust
Inlet
CH*
Quartz optical
windows Quartz combustor liner
OH*
Bifurcated optical fiber
Figure 22. LDI combustor in the high pressure test rig.
A schematic diagram of the fuel supply and control setup is shown in Figure 23.
Fuel is pressurized and supplied using a constant displacement fuel pump, with a return
circuit for unused fuel. The pump, driven by an electric motor, can pressurize fuel up to
2000 psig. A pressure regulator is used after the pump to reduce the pressure to a required
maximum value. A pneumatic actuated valve is used to control the fuel flow. The control
valve takes a 4-20 mA signal as input and requires an air pressure between 15-20 psig for
actuation. A turbine flow meter (OMEGA Engineering Inc, Model: FTB9502) is used for
measuring fuel flow rate. The output of the flow meter is a pulsed signal, with pulse rate
non-linearly proportional to flow rate. Hence a linearizing flow computer (OMEGA
Engineering Inc, Model: FC-21) is used to convert the non-linear output from the flow
meter to a voltage signal that is linearly proportional to flow rate. The flow meter was
calibrated for measuring Jet-A by direct measurement of the total fuel volume for a
known flow time. In addition to the primary turbine flow meter, the fuel line pressure
before the nozzle is also measured in order to provide a second estimate of fuel flow rate
(see following paragraph). A solenoid valve located just upstream of the nozzle permits
rapid on/off control of the fuel flow. To prevent fuel coking due to residual fuel in the
43
nozzle and supply tube, a fuel purge system using nitrogen gas is employed. Residual
fuel is purged out immediately after fuel to the combustor is turned off. Air mass flow
rate through the combustor is measured using a differential pressure orifice meter.
Combustor operating pressure and inlet air temperature are measured using transducers
mounted on the inlet section of the test rig, just before the combustor.
Pressure
transducer Solenoid Valve
Filter Pressure
regulator
Solenoi
N2 d valve
Accumulator N2 cylinder
Check
Pressure Valve
Fuel transducer
pump
To fuel
Pneumatically Turbine Solenoid
Filter nozzle
actuated flow flow meter valve
control valve
Figure 23. Schematic diagram of fuel supply and control setup.
The fuel nozzle was calibrated to obtain the relationship between its pressure drop
and the Jet-A mass flow rate, or essentially its Flow Number (FN) as defined in Eq. (3.1).
The experimental results, along with a linear fit, are shown in Figure 24. As expected for
an incompressible flow, the flow rate is linearly proportional to the square root of
pressure difference across the nozzle. The slope gives FN, with a value of 0.683. This
information was used to choose appropriate settings for fuel supply line pressure, and to
support the flow rate measured by the turbine flow meter.
44
m (lb/hr)
FN  (3.1)
 P(psi)  0.5
16
Jet-A mass flow rate (lb/hr) 14
m  0.683 P
12
10
8 Experiment
6 Linear fit
2
5 10 15 20 25
Square root of nozzle Pressure differential(psi0.5)
Figure 24. Pressure dependence of fuel nozzle mass flow rate for Jet-A.
The average velocity in the combustor (the quartz tube) in the absence of
combustion was in the range 9-15 m/s to match practical engine combustor conditions.
For example, combustor velocity is limited to prevent excessive pressure loss due to heat
addition, which increases as a function of square of combustor reference velocity [7].
Based on the typical fuel flow rates (1.0-1.76 g/s) employed, the thermal loading of the
combustor was between 28 and 74 kW. While the fuel and air flow rates were measured,
the combustor equivalence ratio is difficult to define. It is possible that some part of the
fuel spray from the central injection region penetrates into the co-flow and burns there.
The amount of this fuel cannot be determined easily and the penetration into the co-flow
likely varies with operating conditions, e.g., fuel nozzle pressure differential and
combustor pressure. Hence, the primary zone equivalence ratios is not known. Therefore
only overall equivalence ratios, based on the total combustor air flow, are used here to
describe the operating conditions. This is acceptable for the present study, as the actual
45
flame equivalence ratios are not important; only the combustor’s operating stability
margin (Φ/ΦLBO) is relevant.
The single element LDI injector is intended to produce a lean, partially premixed
flame like the original nine element NASA LDI injector in order to produce low
emissions. In addition, when operating as desired, the single element injector should
qualitatively reproduce the shape and size of the flame associated with each element of
the multi-element injector. As the NASA single-element injector design had not been
tested previously, considerable amount of effort was made to test and develop the single
element LDI injector in order to meet the performance goals. The efforts were focused on
modifying: the flow split ratio between the perforated plate and the LDI element, the co-
flow velocity profile, and the fuel nozzle position relative to the venturi throat. More
details of the combustor development are given in Appendix C.
The LDI injector is more likely to produce a nearly premixed flame at higher
operating pressures where higher fuel flow rates are required, and thus the pressure
differential across the fuel nozzle is higher, which results in a finer fuel spray. At
relatively lower pressures, and near LBO, the flame might not be as well premixed due to
lower pressure differentials across the fuel nozzle. However, the current conditions
represent low power, part load operation where LBO is likely an issue.
Flame images obtained with a digital color camera at the end of development
process are shown in Figure 25, for combustion operation at atmospheric pressure and
elevated pressures. The images show a blue flame indicating low emission operation. In
addition, the flame is quite short in length with slightly higher width then the LDI
element diameter, indicating adequate confinement by the co-flow.
46
(a) (b) (c)
Figure 25. Flame in the LDI combustor at (a) p=1 atm, v=14 m/s, T= 663 K ,
Φoverall=0.41; (b) p=2 atm, v=10.6 m/s, T=682K, Φoverall=0.28; and (c) p=4 atm,
v=12 m/s, T=733 K, Φoverall=0.23 .
3.2.2. LBO Sensing and Diagnostics
A bifurcated optical fiber with an acceptance cone angle of 32 is used for
collection of the optical radiation from the combustor. The fiber is placed nominally at an
angle of 45° with respect to the combustor axis, viewing most of the flame (see Figure
22). The collected light is split into two parts by the bifurcated fiber: one part passes
through an interference filter centered at 308 nm corresponding to OH* emission and the
other through a 419 nm filter corresponding CH* emission. The same type of
photomultipliers used in the gas-fueled setup is also used here to detect the optical
signals.
Acoustic radiation from the flame is monitored with a Kistler piezo-electric
pressure transducer. The transducer is side-mounted onto a long (~10 m) standoff tube
connected to the combustor.
High-speed flame images are acquired with a CMOS camera (Photron,
FASTCAM) fitted with an optical band-pass filter that transmits over a range of ~320-
600 nm. A Schott glass (BG 28) filter with a transmission range of ~325-600 nm was
47
placed in front of the camera lens. The camera views the combustor from a downstream
direction, at an angle of about 45° similar to the optical fiber. Images are acquired at a
rate of 1000 fps and an exposure of 1 ms, and are synchronized with the optical and
acoustic sensor data recording.
3.3. Turbofan Engine Model and Control
In engines, LBO could occur during rapid power reduction (deceleration)
transients. When fuel is decreased sharply to reduce power, air flow through the engine
drops rather slowly, as compressor speed slowly decreases due to inertia of the rotating
turbomachinery. This results in lower combustor fuel air ratios, which could result in
flame blowout. To prevent LBO during deceleration transients, current aircraft engine
control systems commonly use a passive method which puts a minimum allowed limit on
Ratio Unit (RU), defined as Wf/Ps3, where Wf is fuel flow rate and Ps3 is combustor inlet
static pressure. Due to uncertainly in the LBO limit, the RU limiter is much higher than
the actual RU limit where LBO would occur, which limits the transient response. An
active LBO margin sensor based on precursor event detection would allow using lower
RU limiter, thus improving the transient response.
In order to investigate the possible improvements in transient response along with
the limitations in implementing precursor event based active LBO margin sensors, a high
bypass turbofan engine, similar to commercial aero engines, is modeled using the
Numerical Propulsion System Simulation (NPSS) package. The engine model is similar
to the model used in NASA engine control simulation software C-MAPSS40K [76, 78].
The engine model also includes a controller designed for full power to idle transient at
sea level static conditions.
3.3.1. High Bypass Turbofan Engine Model
The engine model is a high bypass ratio turbofan engine with two spools
producing a maximum thrust of 35,000 lbs at standard sea level static conditions. A
48
schematic of the engine along with the flow paths is shown in Figure 26. The engine
consists of six main components; fan, low pressure compressor (LPC), high pressure
compressor (HPC), combustor, high pressure turbine (HPT) and low pressure turbine
(LPT). In addition, it consists of an inlet, bypass duct, nozzle and various cooling bleeds.
The HPT and HPC are connected through a high speed shaft, whereas the LPT, LPC and
the fan are connected through a low speed shaft (fan shaft). For modeling purposes, the
fan hub and the LPC are combined together into one unit where the fan hub acts as an
additional stage for the LPC. Flow through the fan tip passes directly in to the bypass
duct.
Wf
Ps3 Nozzle
HPC HPT
LPC LPT
Fan
Figure 26. Schematic of the simulated turbofan engine with flow paths.
The engine model is implemented in NPSS. The model is a zero-dimensional,
physics-based component level model. The model essentially computes output properties
of each component using inputs (pressure and temperature) at a given a mass flow rate.
The five turbomachinery components are characterized by their maps relating pressure
ratio, corrected mass flow rate and adiabatic efficiency, which are parameterized with
corrected shaft speeds. The combustor is characterized by its combustion efficiency and
49
pressure drop. Various flow ducts are characterized by their pressure losses. The model
requires design point specifications for calculating engine performance at off-design
operation.
The design point for the engine is considered to be full power operation at sea
level static conditions (Altitude=0, Mach number=0). The specifications for various
components at the design point are given in Table 1. Maps for the five turbo machinery
elements are taken from the example maps provided with NPSS.
Table 1. Design point specifications of the turbofan engine.
Component Parameter Value

Fan Pressure ratio 1.66
Bypass ratio 5.53
Efficiency 0.86
Low-pressure compressor Pressure ratio 2.37
Efficiency 0.87
High-pressure compressor Pressure ratio 12.22
Efficiency 0.85
High pressure turbine Pressure ratio 4.42
Efficiency 0.87
Low pressure turbine Pressure ratio 3.84
Efficiency 0.89
Combustor Fuel flow rate 3.62 lbm/s
Efficiency 0.98
Low-speed shaft (fan shaft) Inertia 106 lbf-ft2
Rotational speed 4039 rev/min
High-speed shaft Inertia 21.5 lbf-ft2
Rotational speed 11868 rev/min
Inlet Airflow rate 104 lbm/s
3.3.2. Engine Control
The engine controller is primarily designed for thrust reduction from full power to
idle, and implemented in NPSS. Power lever angle (PLA) position (80°: full power - 40°:
50
idle) is given as an input to the controller and it is converted to fan shaft speed demand.
The PLA position is mapped linearly to thrust, whereas it is non-linearly mapped to the
shaft speed, as the shaft speed and thrust are not linearly related. The fan shaft rpm is
controlled to achieve a desired thrust by varying the fuel flow rate. The control system
architecture is shown in Figure 27. The system consists of a proportional-integral (PI)
controller to achieve desired shaft speed. The PI controller takes error in shaft speed as
input and outputs the required change in RU. RU is used as the control variable as it is
the common control variable in engines. RU is multiplied by the current value of Ps3 to
obtain the required fuel flow rate. To prevent LBO during deceleration, an RU limiter is
employed similar to conventional engine control systems. The limiter is a constant and
not scheduled. The maximum of the fuel flow rates given by the PI controller and the RU
limiter is commanded to the fuel actuator.
Ps3 Wf
RU limiter
Max
Ps3
PLA ΔRU Nout
Ndmd e PI Gain RU Wf Actuator Wf Engine
+ + +
-
Steady state RU
Figure 27. Architecture of the engine control system.
Turbine engines are highly non-linear systems from control perspective, as they
behave differently at different operating points. However they can be considered linear in
a limited range around an operating point [77]. This allows a set of simplistic linear
controllers (e.g., PI) to be used for controlling non-linear engine systems. The control
usually requires scheduling and interpolation of control gains each set tuned for a specific
51
operating point. Since only a rapid transient is considered, only two sets of gains, one
near full power and the other near idle, were found to be adequate for the current study.
The PI control gains were tuned such that rapid time response is achieved without
considerable overshoot. This was achieved by obtaining a transfer function from RU to
shaft speed using system identification process in MATLAB. A chirped fuel input signal
was given to the engine and its shaft speed response was obtained in order to estimate the
transfer function. The transfer function is assumed to be second order and given by Eq.
(3.3). The parameter values of the transfer function are given in Table 2. Using the engine
transfer function, PI controller gains were tuned in MATLAB to produce optimal control
response. The gains at full power are Kp=6.2 lb/s/psi, Ki=12.3 lb/s/psi/s and at idle
Kp=23.5 lb/s/psi, Ki=9.62 lb/s/psi/s, where Kp is the proportional gain and Ki is the
integral gain (Eq. (3.2)). The actuator transfer function is assumed to be the first order
response given by Eq. (3.4) with Ke=1 and Tp1=0.02 s.
t
U  K p e  t   Ki  e  d (3.2)
0
Ke (1  Tz s)
G( s )  (3.3)
1  Tp1s 1  Tp2s 
Ke
A( s)  (3.4)
1  Tp1s
Table 2. Engine transfer function parameters at full power and idle
Operating point Ke (rpm/lb/s/psi*10-3) Tz(s) Tp1(s) Tp2 (s)

Full power 0.117 0.391 1.03 0.535
Idle 0.12 0.362 1.83 0.910
52
During the deceleration transient control when the RU limit is reached, control
switches from the PI controller to the RU limiter. Since the PI control would be inactive
when the limiter is on, the integral error term keeps accumulating due to the integration
action, and is known as an integral windup problem [77]. The integral windup results in
excessively large outputs which fails the PI controller. Integral windup protection is
usually employed to reduce the integral error when the PI controller is not active. In the
present controller, windup protection was implemented by modifying the integral error
term using Eq. (3.6), where ew is the error between RU commanded to the engine and RU
calculated by the PI controller (3.5). A value of 3.0 lb/s/psi/s for Kw was observed to
produce the desired results.
ew   RU out  RU PI  (3.5)
t
 K 
intErr    e  w ew dt (3.6)
0
Ki 
3.3.3. Example Control Results
Example control results for full power to idle transient, initiated at t=1 s by
moving the throttle angle from full power (80°) to idle (40°) are shown in Figure 28. The
time step between successive simulation runs was 10 ms. The RU limiter is taken to be
15.12 lb/hr/psi, which is slightly below the steady state idle RU of 17.6 lb/hr/psi. From
the thrust response, it can be seen that the control produces a smooth transition to idle
without any undershoot, taking up to 7 s to reach idle.
The controller decreases fuel flow rapidly at the beginning and thereafter slowly
decreases it, governed by the rate at which the air flow rate (i.e., Ps3) drops. From the RU
limiter on/off plot, the limiter comes on within 150 ms after the transient initiation and
stays on for 4.7 s. Therefore most of the deceleration is achieved while keeping the RU at
the minimum limit. From the RU plot, RU is maintained constant during the time when
53
RU limiter is on, and starts to increase when the engine is sufficiently slowed down in
order to match the final desired speed. The equivalence ratio is calculated by assuming
27% of the flow through the combustor passes through the primary zone.
This well-behaved simulation of the aeroengine, including the controller, is used
to provide the basis on which the LBO margin sensing approach is tested for its
applicability to control of rapid engine transients in Chapter 6.
RU limiter
80 1
PLA
Throttle (deg)
RU limiter
60 0.5
40 0
0 4
5 10 0 5 10
x 10
Thrust 30 RU
Ratio Unit (lb/hr/psi)
3
Thrust (lbf)
25
2
20
1
15
0 5 10 0 5 10
4

Fuel flow rate (lb/s)
Wf
3 1.5
Equivalence Ratio
2
1
1
0 0.5
0 5 10 0 5 10
Time (s) Time (s)
Figure 28. An example full power to idle transient starting at t=1 sec.
54
CHAPTER 4
LBO MARGIN SENSING: OPTICAL DETECTION
This chapter details results for LBO margin sensing based on analysis of optical
emissions. The margin sensing approach is mainly by detection of flame partial
extinction and re-ignition events occurring near LBO in flame chemiluminescence
signals. Results are presented for combustors operating under dynamically unstable
conditions, and at elevated pressure and temperature. Section 4.1 covers LBO margin
sensing under dynamically unstable conditions in a gas-fueled combustor. Section 4.2
describes margin sensing at elevated pressure and inlet temperature in a liquid-fueled LDI
combustor. The final section summarizes the results.
4.1. LBO Margin Sensing in the Gas-Fueled Combustor
LBO margin sensing under high amplitude dynamic instability is investigated in a
gas fueled combustor that can produce instability through at least two distinct
mechanisms: instability without equivalence ratio oscillations (w/o Φ′) and instability
with equivalence ratio oscillations (w/ Φ′). Heat release oscillations for instability w/o Φ′
are mainly due to flame-vortex interactions, whereas both flame-vortex interactions and
equivalence ratio oscillations contribute to heat release oscillations in the case of
instability w/ Φ′. The study was focused on these two mechanisms, as they are the two
main instability mechanisms in lean premixed combustors [72]. Besides the dynamically
unstable conditions, results from dynamically stable operation are also presented in order
to compare them with the unstable conditions. The combustor is described in Chapter 3.
For the dynamically unstable experiments all operating conditions were identical, except
for the fuel injection location. For examining combustor behavior as it approaches LBO,
air flow rate was kept constant and fuel flow rate was reduced. Average cold flow
55
velocities were 4.2 m/s and the equivalence ratios where blowout occurred were: 0.688
for dynamically stable operation, 0.678 for instability w/o Φ′, and 0.694 for instability w/
Φ′. OH* chemiluminescence signals were recorded for a period of 30 s at each operating
condition.
4.1.1. Instability Characteristics
In order to study LBO margin sensing under dynamically unstable conditions, a
significant level of dynamic instability in lean conditions including near LBO is required.
The current combustor meets this requirement by exhibiting pronounced dynamic
instability in lean conditions, over a wide range of equivalence ratios, including very
close to blowout, for both kinds of instability mechanisms. The dynamic instability of the
combustor is characterized in Figure 29, which shows acoustic signals recorded by a
microphone outside the combustor. Signals are shown for the combustor operating near
LBO for each of the fueling cases (w/ and w/o Φ′). For the two cases, all combustor
operating conditions are identical except for the fuel injection location. Both signals
clearly exhibit periodic oscillations without much noise, illustrating the existence of
pronounced dynamic instability near LBO. In addition, the sound pressure level outside
the combustor at the microphone location is about 130dB. Though not measured, the
pressure level inside the combustor would be much higher. A clear difference between
the two traces is the presence of strong amplitude modulations for instability w/ Φ′. In
addition, the dynamic instability is more pronounced w/ Φ′, as indicated by the higher
peak-to-peak amplitudes compared to the w/o Φ′ trace.
Power spectra corresponding to the acoustic signals of Figure 29 are shown in
Figure 30. The spectra are obtained by ensemble averaging multiple spectra each having
a resolution of 0.3 Hz (data were recorded for 30 sec and broken into 3 sec segments).
Most of the acoustic power emitted is emitted in a relatively narrow region, with very low
power due to broadband combustion noise. The instability peaks are located at 265 Hz
56
(w/o Φ') and 245 Hz (w/ Φ'). These frequencies are close to the axial quarter wave
acoustic mode frequency of the combustor. Assuming a bulk uniform product
temperature of 1000 K, the quarter wave mode frequency is 263 Hz. The instability
frequencies are slightly different and in addition, instability w/ Φ' has a broader peak,
which corresponds to the significant amplitude modulation seen for that case. In addition
the broad peak suggests that the amplitude modulations are not periodic. If they were
periodic, a clear second peak in the spectrum should be evident.
5 w/o '
Amplitude (a.u.)
-5
0 50 100 150
5 w/ '
-5
0 50 100 150
Time (ms)
Figure 29. Acoustic signal time traces for combustor operation near LBO, at
Φ=0.71 for instability w/o Φ′ Φ=0.72 for instability w/ Φ′.
0.12
0.8 W/O  '

W/  ' 0.1
0.6 0.08
0.06
0.4
0.04
0.2
0.02
0 0
0 100 200 300 400 500
Frequency (Hz)
Figure 30. Power spectra of acoustic signals for combustor operation near LBO at
Φ=0.71 for instability w/o Φ′ Φ=0.72 for instability w/ Φ′.
57
Based on these results, it can be concluded that the instability characteristics are
noticeably different for the two fueling methods. The disparities provide evidence that
fueling the combustor just upstream of the final swirler does add complexity and
modifies the instability mechanism, adding equivalence ratio variations.
4.1.2. LBO Precursor Events
In previous LBO approach detection studies in a premixed swirl combustor, the
flame was observed to exhibit partial extinction and re-ignition events near LBO. These
events, called LBO precursor events, were observed to occur more often as the
combustor’s stability margin (Φ-ΦLBO) was reduced, thus providing a measure of
combustor’s proximity to LBO. The LBO approach detection with event identification
provides better time response compared to other approaches such as low frequency
spectral content estimation or standard deviation estimation, due to finite time required
for their evaluation. Therefore the present study focused mainly on the LBO precursor
event identification approach. As described in Chapter 2, precursor events in optical
(chemiluminescence) signals are characterized by a drop in signal amplitude from the
time of extinction to the time of re-ignition.
Before examining optical LBO precursor events under dynamically unstable
conditions, an example precursor event under dynamically stable conditions is shown in
Figure 31. The combustor was made dynamically stable by using a shorter tube with a
length of 300 mm instead of the standard 600 mm tube. Shortening the tube length
increases the natural frequency of the (longitudinal) acoustic mode, to ~530 Hz. For this
case, fuel was injected far upstream. The resulting chemiluminescence signal shown in
Figure 31 was acquired from a region of about the combustor diameter in length above
the combustor inlet. The signal does not exhibit sinusoidal oscillations. There is evidence
of some low amplitude, quasi-periodic fluctuations with a period of ~25 ms. The
amplitude is only a few times greater than the level of the combustion noise. However,
58
the corresponding frequency, 40 Hz, does not coincide with the natural frequency of the
combustor based on the combustor’s length. Based on these observations, we denote this
mode of operation as dynamically stable, especially in comparison to the other cases
where dynamics were very strong. The local mean amplitude of the signal starts to drop
at around 100 ms and takes about 50 ms to recover. This feature is similar to precursor
events observed in earlier studies, under dynamically stable conditions, in a premixed
swirl combustor with a slightly different geometry [69].
3
Amplitude (a.u.)
0
0 50 100 150 200 250
Time (ms)
Figure 31. OH* chemiluminescence signal time trace with a precursor event
occurring between 100-150 ms for dynamically stable operationat Φ=0.704.
Figure 32 shows OH* signal time traces during dynamically unstable operation,
near LBO, for instability w/o Φ' and w/ Φ'. For instability w/o Φ', there are usually large
amplitude modulations due to the instability, and the minima of the oscillations are well
above zero. Between 130-160 ms, there is a precursor-like feature characterized by a
large drop in signal amplitude for a duration of about 30 ms. During this period,
amplitude modulations are diminished and resume only after complete re-ignition, i.e.,
after the apparent precursor event has ended. The signal behavior during this period is
somewhat similar to the precursors in dynamically stable conditions. Also this type of
characteristic signal feature was observed only when the combustor was operating near
LBO, indicating that they are indeed LBO precursors.
59
3 w/o '
1
Amplitude (a.u.)
0
0 50 100 150 200 250
4 w/ '
0
0 50 100 150 200 250
Time (ms)
Figure 32. OH* signal time traces with precursor events for dynamically unstable
operation w/o Φ' and w/ Φ'. Mean equivalence ratios: 0.698 (w/o Φ'), 0.714 (w/ Φ').
Unlike the w/o Φ' mode, instability w/ Φ' has amplitudes dropping to zero during
each instability cycle (see Figure 32), independent of whether the combustor is operating
near LBO. Thus low amplitudes (optical signals near zero) alone cannot be taken as an
indication of proximity to LBO. However there are other instances in the signal that
could be precursor events. For example between 125-150 ms, there are no amplitude
modulations, and the local signal mean stays well below the normal (long term) mean of
the signal. Such events were observed only close to LBO, indicating that they are
precursors. The event has a duration of about 25 ms, similar in time scale to the precursor
for instability w/o Φ'. As the two events look qualitatively different, the event
identification analysis would require different strategies for the two modes of operation.
High speed flame chemiluminescence imaging was carried out in order to
examine the flame behavior near LBO during precursor events. High speed images
during combustor operation near LBO, for instability w/o Φ' are shown in Figure 33.
Images were captured at 1900 frames/s with an exposure time of 200 µs for each frame.
Grayscale images were scaled with intensity based color maps, as indicated in the figure.
60
An outline of the combustor is superimposed on each image. The images are broadband
chemiluminescence images without any spectral filtering. The images in the top row of
the figure correspond to normal combustor operation, while the bottom row images
correspond to a precursor event, similar to the one in Figure 32. Both image sequences
are from the same run, with the bottom row sequence starting 200 ms after the end of top
row sequence.
Figure 33. High speed flame chemiluminescence images for instability w/o Φ', Φ
=0.698. (Top row) Images during normal combustion over an instability cycle with
an image separation of 0.526 ms. (Bottom row) Images during a precursor event,
every 13th image shown (effective image separation of ~6.8 ms). In the color bar red,
represents the highest chemiluminescence emission intensity and blue represents the
lowest.
Normal combustor operation (top row images) is characterized by instability
cycles, and the top row images show roughly one full cycle, starting from a peak and
ending at a peak in a heat release oscillation cycle. The flame length and radiation
intensity are clearly modulated over a cycle. Bright and short flames correspond to high
heat release, whereas long and weaker (lower radiation intensity) flames correspond to
61
low heat release in a cycle. High speed images for the combustor operating at
equivalence ratios far from LBO exhibit similar periodic instability cycles.
In the bottom row of Figure 33, which corresponds to a precursor event, large
scale flame extinction around the inner recirculation zone is seen, followed by its
recovery. From other frames in the sequence, the normal flame was observed to
extinguish gradually followed by a gradual recovery. Most of the time during the
extinction period, there is a small region of heat release just above the center body.
However on occasion, this center body flame was also lost, as seen in the 4th frame.
During the period of extinction, the flame switches to a lifted configuration, where most
of the burning occurs downstream. The stable flame around the inner recirculation zone
and the temporarily lifted flame can be considered two different flame modes. Precursor
events in this combustor are associated with flame switching between these two modes.
Similar flame shifting to lifted mode was observed in prior LBO margin sensing studies
on a geometrically similar, gas-fueled combustor, under dynamically stable conditions
[69].
High-speed flame chemiluminescence images during normal combustion and a
precursor event, for instability w/ Φ' are shown in Figure 34. As before, both image
sequences are from the same run. The images were acquired at 1000 frames/s with 500 µs
exposures. In the images, a region of about 25 mm above the combustor was not
optically accessible, unlike the images in Figure 33. The precursor event image sequence
(bottom row) starts about 1.4 s after the end of the normal combustion image sequence.
62
Figure 34. High speed flame chemiluminescence images for instability w/ Φ',
average Φ =0.711. (Top row) Images during normal combustor operation with the
first four images during a high amplitude instability cycle, separated by 1 ms; last
two frames during a period of low instability. (Bottom row) Images during a
precursor event with an image separation of 6 ms.
In contrast to the w/o Φ' case, normal combustor operation w/ Φ' is characterized
by unsteady instability cycles, i.e., regular growth and decay in instability amplitude.
Figure 35 displays an optical signal time trace (from the fiber optic sensor) acquired
simultaneously with the high speed images during normal operation (top row in Figure
34). In Figure 35, the times corresponding to each high-speed image are indicated by
circles. The first five images in the top row are during a high amplitude instability cycle,
and the next two are during a period of low instability. During instability cycles, flame
length and intensity modulations are evident, similar to instability w/o Φ'. However, the
modulations in intensity are much larger (the same image intensity scaling is used in the
figure for both cases), suggesting that equivalence ratio oscillations are significant.
63
Higher equivalence ratios during an oscillation cycle would make the flame brighter,
whereas low equivalence ratios make it weaker. During the period of reduced instability
(last two frames), the flame is longer on average, with small oscillations in intensity and
length. From the 2nd and 3rd images, during troughs of an instability cycle, some amount
of flame extinction is observed. This apparent extinction along with a weak flame results
in the optical signal amplitude dropping close to zero during the troughs of the instability
cycle. Though partial extinction is occurring during these instability cycles, they should
not be considered LBO precursors as the combustor shows no tendency to blowout and
can operate continuously. Such statically stable operation could result from the instability
amplitude decaying after its initial growth, thus leading to a more stable flame as seen in
the last two frames. If the instability amplitude were continue to remain the same, or even
rise, it might lead to complete flame loss and blowout.
In contrast to the short duration extinctions, the bottom row in Figure 34 contains
a sequence of images with extinction occurring for a much longer duration (~23 ms) with
more pronounced extinction. The extinction is around the inner recirculation zone,
similar to the extinction during precursors for instability w/o Φ'. Such long duration
extinctions were found to occur only near LBO, and hence can be considered precursors.
Amplitude (a.u.)
1 .5
0 .5
0
0 10 20 30 40
Time (ms)
Figure 35. Optical signal time trace acquired simultaneously with the high-speed
images in Figure 34 (top row images), for instability w/ Φ'. Image locations are
indicated by circles.
64
4.1.3. Precursor event detection
For robust LBO margin sensing, precursors should be detected reliably, i.e.,
without missing events and not detecting false events. Precursor events can be detected
using the double thresholding method used in earlier work, along with a time constraint.
The method consists of a lower and upper threshold evaluated using recent signal
statistical properties, signal mean () and standard deviation (). An example precursor
event along with the thresholds is shown in Figure 36. The lower threshold is set
sufficiently below the minima of the nominal signal to detect the amplitude drop during a
precursor event. The upper threshold is set close to the signal mean such that noise or
instability during a precursor does not cross this threshold, which otherwise would result
in a single event counted multiple times. An event starts when the signal level drops
below the lower threshold and ends when it rises above the upper threshold. A minimum
time constraint can also be imposed on the duration a precursor spends between the two
thresholds for more robust event detection. Both the lower and upper thresholds are
determined from the relation -, where  is a constant. The upper threshold is set with
a smaller value of  compared to the lower threshold.
Upper threshold
Amplitude
Lower threshold Time Duration
Figure 36. An example precursor event in optical signal along with the thresholds
used for its detection.
65
As noted above, precursor events during dynamically stable conditions and during
instability w/o Φ' look qualitatively similar. The signal amplitude drops below the
nominal signal minima during a precursor. However, the nominal signal minima are
considerably lower during the instability due to high coherent modulations in the signal.
The double thresholding method could be used for event identification in both the cases;
however, it would require different  values. For example, if the same  were used for
instability w/o Φ' as for the dynamically stable case, it would result in the lower threshold
having negative values, which is not helpful. Moreover, precursors for instability w/ Φ'
look qualitatively different from the other two cases; the signal amplitude does not drop
below the nominal signal minima, because the nominal signal minima are very small,
close to zero. Hence, the lower threshold, used for detection of amplitude drop below the
nominal signal minima, is not useful. However, the events could be detected in other
ways, for example, when the signal amplitude stays below the signal mean for more than
a specified duration. In summary, the three cases (dynamically stable, instability w/ Φ'
and w/o Φ') would require different event identification algorithms to analyze the raw
signals.
In practical combustors, operation can switch between dynamically stable and
unstable operation. In addition, the dynamic instability mechanism, frequency and
amplitude may change with operating conditions. These changes would require using
multiple LBO precursor identification algorithms and algorithm switching based on the
status of the dynamic instability e.g., amplitude and frequency. However, such an
elaborate approach requires careful development and validation, and may fail to work
reliably. A single algorithm that could work regardless of the status of dynamic instability
is desirable for robust LBO margin sensing. It can be speculated that getting rid of the
instability component, as it is not associated with precursor events, could make a single
event detection algorithm work robustly under all conditions.
66
w/o '
-2 w/ '
10
A.S.D. (a.u.)
-3
10
1 10 100 250 1000

Frequency (Hz)
Figure 37. Amplitude spectra of optical signals for instability w/ Φ' and w/o Φ'.
Equivalence ratios: 0.698 (w/o Φ'), 0.714 (w/ Φ').
From precursors in the optical signals, the duration of the events in this combustor
are between 30-50 ms. On the other hand, the dynamic instability period for the
combustor is about 4 ms, much shorter than the precursor duration. Amplitude spectra of
the optical signals, for the time series data used in Figure 32, are shown in Figure 37 for
both instability cases. The spectra are ensemble averaged over several individual spectra.
Besides the instability peak around 250 Hz, the amplitude spectral density is nearly
constant in the range 1-30 Hz for instability w/o Φ' and 1-50 Hz for instability w/ Φ' and
starts to drop before rising near the instability frequency. The relatively high power in the
low frequency range (1-50 Hz) is most likely due to precursor events due to their longer
time scales. Since the spectral content of the events and the instability are widely
separated, signal processing techniques such as low-pass filtering can be used for
suppressing the dynamic instability.
67
1.5
Amplitude (a.u.)
1
0.5 No instability
instability w/o '
instability w/ '
0
0 50 100 150 200 250
Time (ms)
Figure 38. Precursor events in low-pass filtered (50 Hz) optical signals for
dynamically stable and unstable w/o Φ' and w/ Φ' conditions.
Low-pass filtered optical signals during precursor events for dynamically stable
and unstable conditions are shown in Figure 38. These filtered time traces correspond to
the raw traces shown in Figure 31 and Figure 32. Low-pass filtering was performed at
50 Hz with a 4-th order digital Butterworth filter. The cutoff frequency was chosen based
on the observation precursors have spectral content up to 50 Hz in the w/ Φ' instability
case. In addition, due to the slow roll off of the Butterworth filter, the cutoff frequency
should be sufficiently below the instability frequency. However, cutoff frequencies
anywhere in the range 20-100 Hz have been observed to work equally well. The filter is
an infinite impulse response filter, with the group delay, i.e., time delay, a function of
the frequency. In this case the delay is about 8 ms in the pass band, and goes to a
maximum of 12 ms at 110 Hz. For the precursors, the filtering causes a delay of about
9 ms. Several other filter types or different filter orders have been observed to produce
similar results. After filtering, all three precursors have a similar characteristic shape of
slow amplitude drop and recovery.
68
3
0.1 Exp, =0.72
2.5 -4 -3 Gaussian, =0.72
Exp, =0.70
2 0.05 Gaussian, =0.70
PDF
1.5
0
0 .25 .44 0.55
1
0.5
0
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Amplitude (a.u.)
Figure 39. PDFs of low-pass filtered optical signals along with corresponding
Gaussian PDFs for far from LBO and near LBO operation for instability w/o Φ'.
An inset plot with reduced X and Y scales for near LBO case is also shown.
Such similarity should enable use of a single event identification algorithm, such
as the double thresholding approach, over the full range of operating conditions,
regardless of the dynamic instability characteristics. The lower threshold selection for
detection of precursor events should be based on the signal’s noise characteristics and the
amplitude drop during precursor events. To help in selection of this threshold setting,
probability density functions (PDF) of the low-pass filtered optical signals at two
equivalence ratios, along with the corresponding Gaussian PDFs for instability w/o Φ' are
shown in Figure 39. The Gaussian PDFs are calculated using the mean and standard
deviation of the corresponding filtered optical signals. The signal PDF closely matches
the Gaussian PDF at operation far from LBO (Φ=0.72). This implies that a lower
threshold below -3 has a very small probability of signal dropping below the threshold
under normal operation, and such a threshold would produce only a few false events.
Near LBO, the PDF somewhat deviates from the Gaussian, having a smaller width with a
higher probability for low amplitudes. This behavior can be attributed to precursor
events. For the near LBO PDF, an enlarged version is also shown in the inset plot with X
69
and Y axis scales reduced. The probability starts to increase below -3 and peaks at -
4. Therefore, the lower threshold setting for precursor detection should be at least below
-3, whereas -4 may work better. Similar results were observed for instability w/ Φ'.
4.1.4. LBO Margin Detection
Average event occurrence rates obtained with different lower threshold settings in
the range -3 to -3.75 for instability w/o Φ' are shown in Figure 40. The upper
threshold and the minimum duration constraint are kept at -1 and 10 ms respectively.
The threshold level at a given time is calculated from signal data over the previous one
second period. The results are from low-pass filtered optical signals at 50 Hz and are
averaged over 29 s at each equivalence ratio. In evaluating the  and  for the threshold
calculation, signal data during precursor events are omitted as it has been observed to
lower the thresholds if large duration precursors or more precursors occur in the
evaluation time window. In addition, any deviation above and below the -3 limit is
also excluded. From the figure, all the threshold settings produce low event rates far from
LBO and high event rates near LBO. This suggests the robustness of event detection to
moderate changes in threshold setting. However, an optimal threshold setting should
produce almost no events far from LBO, to minimize false alarms. On the other hand,
events should be the maximum possible near LBO (detection of all events) for a better
indication of LBO margin. The thresholds, -3.0, -3.25 and -3.5 produce
somewhat non-zero events far from LBO whereas -3.75 produces zero events far from
LBO. Therefore a lower threshold of -3.75 is an appropriate choice for event
identification.
Since the precursor event durations are sufficiently large, they are well above the
nominal noise time scales. Hence event sensing has been observed to be not sensitive to
70
the duration constraint. In addition, as the upper threshold is mainly used for imposing
the duration constraint, it is not independent from the duration constraint.
Equivalence Ratio (Φ)

0.68 0.7 0.72 0.74 0.76
2.5
 -3.0
Average number of events/s
 -3.25
2
 -3.5
 -3.75
1.5
0.5
0
0 0.02 0.04 0.06 0.08 0.1
Stability margin (Φ- ΦLBO)
Figure 40. Average number of events per second for different lower threshold
settings for instability w/o Φ'. Upper threshold=-1 and minimum time
constraint=10 ms.
Results of event identification for the three operational modes of the combustor,
i.e., dynamically stable, unstable w/o Φ' and w/ Φ', are shown in Figure 41. Similar to the
above results, the events are obtained from low-pass filleted optical signals (@ 50 Hz)
and are averaged over 29 s. The optimal threshold settings for instability w/o Φ', lower
threshold of -3.75 upper threshold of -1.0 and a minimum time constraint of 10 ms,
are also used for the other two cases. Similar to the observations in previous work, the
average event occurrence rate increases as the combustor approaches the LBO limit.
Therefore, the number of events occurring per unit time is a reasonable measure of
proximity to LBO. A single event identification algorithm produced an acceptable event
rate trend for all the three cases. This further demonstrates the ability of low-pass filtering
to allow for one, universal event identification algorithm to be successful regardless of
71
the state of dynamic instability. As shown in the figure, events start to occur around a
stability margin of 0.03 (early warning margin) and the average event rate is between 0.2-
2 events/s. In addition, the event rate increases somewhat monotonically with reduction
in stability margin. However, it drops slightly just before LBO for both the dynamically
unstable cases. Relatively large duration precursors have been observed to occur just
prior to LBO and the event rate drop can be attributed to this. For both the dynamically
unstable cases, event rates are similar, whereas it is nearly half for the dynamically stable
case. This behavior could be attributed to the increased likelihood for an event to occur
because of large perturbations in velocity and equivalence ratio during instability.
The duration of precursor events is an important feature of the precursors, and in
prior studies it was observed to increase towards LBO. The average duration of a
precursor, obtained by averaging over all the precursors detected at given equivalence
ratio, is shown in Figure 42 for the three operational modes of the combustor. Since low-
pass filtering smoothens signal, the actual duration of precursors cannot be obtained from
the filtered signals. Therefore, unfiltered signals were used to determine the event
durations. For this analysis, the duration is defined as the time spent by the signal below
the local mean during a precursor. The precursor event detection algorithm was different
for the three cases as a single algorithm could not be used for all, due to different
characteristics of the precursors. For the dynamically stable mode, events were detected
using the constraints: signal should spend at least 10 ms between successive crossings of
the mean and during this period the signal level should drop below -3.0 threshold at
some instance. For instability w/o Φ', similar approach was used, with the threshold
changed to -1.25 as the nominal  of the signal is large due to high modulations in the
signal due to instability. For instability w/ Φ', the only constraint used was that the signal
should spend at least 10 ms below the local mean between successive crossings. Each
algorithm was observed to detect precursors reliably and produced similar events rates as
72
the filtered signals. In Figure 42, it can be seen that the average event duration increases
as LBO is approached, for all the three cases. Near LBO, when a considerable number of
events start to occur (stability margin < 0.03), event durations are in the range 15 - 45 ms.
For the w/ Φ' instability case, event durations are considerably lower compared to the w/o
Φ' and dynamically stable modes. It is possible that for instability w/ Φ', the oscillations
create stratification in the equivalence ratio, specifically regions of high equivalence ratio
that could promote re-ignition, thus lowering precursor duration.
2 Dynamically stable
Instability W/O '
Instability W/ '
1.5
0.5
0
0 0.02 0.04 0.06 0.08 0.1
Figure 41. Average event occurrence rates obtained from low-pass filtered optical
signals for dynamically stable, dynamically unstable w/o Φ' and w/ Φ'.
73
50
Average duration of an event (ms)

Dynamically stable
Instability W/O '
40
Instability W/ '
30
20
10
0
0 0.02 0.04 0.06 0.08 0.1
Figure 42. Average event duration obtained from un-filtered optical signals for
dynamically stable, dynamically unstable w/o Φ' and w/ Φ'.
In addition to the event occurrence rate and event duration, the modulation depth
of events might also be expected to increase as LBO approaches, as the the extinctions
are likely to become larger in spatial extent. Therefore, these three signal features
(number of events, duration and modulation depth) can be combined into one single
parameter for robust estimation of LBO margin. The parameter, denoted Stability Index
(SI), is defined in Eq (4.1). Here C(t) is the chemiluminescence signal at time t, C is the
signal local mean, Twindow is the time window used for integration and event is used to
include data only when an event is occurring (1 during events and 0 otherwise). A
discretized form of this expression could be used for SI evaluation from sampled data.
From the definition, SI is non-dimensional, and it is essentially the integrated signal loss
below the mean during events. In physical terms, SI represents the fractional loss in local
heat release due to precursors. This approach weighs the LBO proximity parameter by the
strength of an event, instead of weighing all the events equally as in the use of event
occurrence rate.
74
 C  C (t ) 
Twindow
1
SI 
Twindow 0

 C
dt  event

(4.1)
The Stability Index and the event rate, normalized with corresponding averaged
values in the Φ-ΦLBO range 0.034-0.053, are compared in Figure 43 for instability w/o Φ'.
The SI is calculated from low-pass filtered optical signals at 50 Hz and averaged over 29
s at each equivalence ratio, similar to the event rate. In the plot, an additional vertical-axis
scale (on the right hand side) for the actual values of SI is also shown. SI clearly provides
a higher dynamic range compared to the event rate, and in that sense, a better LBO
proximity parameter. Near LBO, SI varies between 1-7%, suggesting there is a similar
level of reduction in the heat release due to precursors in the region viewed by the sensor.
80 8
Events
70 7
Normalized events and SI
SI
Stability Index (SI) (%)

60 6
50 5
40 4
30 3
20 2
10 1
0 0
0 0.02 0.04 0.06 0.08 0.1
Figure 43. Average event rate and Stability Index (SI) for instability w/o Φ'
obtained from low-pass filtered optical signals. Left vertical axis scale: event rate
and SI normalized with average event rate SI in the range Φ-ΦLBO = 0.03-0.05.
Right axis scale: actual scale for SI.
A similar comparison of SI and occurrence rate for instability w/ Φ' is shown in
Figure 44. Here, event rate and SI are normalized with corresponding values in the Φ-
75
ΦLBO range 0.023-0.044. Again, SI produces a higher dynamic range. Compared to the
results for the w/o Φ' case, instability w/ Φ' has lower values (1-5%), likely due to the
shorter event durations. Though the results presented above ( Figure 40 - Figure 44) are
based on one data set, several other data sets acquired at different times were observed to
produce similar results.
Events
80
5
Stability Index (SI) x 100

SI
Normalized events and SI
60 4
3
40
2
20
1
0 0
0 0.02 0.04 0.06 0.08
Figure 44. Average event rate and Stability Index (SI) for instability w/ Φ' obtained
from low-pass filtered optical signals. Left axis scale: event rate and SI normalized
with average event rate SI in the range Φ-Φ LBO = 0.023-0.044.
4.2. LBO Margin Sensing in the LDI Combustor
In order to investigate the robustness of the LBO margin sensing approach, tests
were also performed in the liquid-fueled, single-injector LDI rig at elevated pressure
(described in Chapter 3). The following results correspond to combustor operation at 2
and 4 atm, with a focus on the 2 atm case. Average un-burnt flow velocity in the
combustor was about ~10 m/s, with ~700K inlet air. To examine the combustor behavior
versus LBO margin, the air flow rate was kept constant and the fuel flow rate was
decreased in steps until LBO occurred. As the exact equivalence ratio of the combustion
could not be determined due to the presence of co-flow (see Chapter 3), results are
76
presented as a function of equivalence ratio normalized by the equivalence ratio where
LBO occurs (Φ/ΦLBO). For 2 atm operation LBO occurred at an overall equivalence ratio
of 0.23 and for 4 atm it occurred at 0.19. Both OH* and CH* chemiluminescence signals
were recorded. In the following results, CH* signals are mainly presented.
4.2.1. Instability Characteristics
2000 2
Acoustic
Power spectral density (Pa2)
Power spectral density (a.u.)

Optical
1500 1.5
1000 1
500 0.5
0 0
500 1000 1500 2000 2500
Frequency (Hz)
Figure 45. Power spectra of CH* signal and acoustic signals for LDI combustor
operation near LBO at Φ/ΦLBO = 1.09; combustor pressure = 2 atm.
The power spectrum of the CH* chemiluminescence signal from the LDI
combustor is shown in Figure 45 for combustor operation at 2 atm near LBO
(Φ/ΦLBO=1.09). For characterization of the dynamic stability of the combustor, a
spectrum is also shown for the acoustic pressure (recorded simultaneously). The spectra
are obtained by ensemble averaging the Fourier transforms of 40 data sets, each having a
spectral resolution of 1 Hz and a Nyquist frequency of 2500 Hz. The LDI combustor
appears to exhibit a moderate level of dynamic instability, as evidenced by the narrow
peaks in the power spectra of the optical (CH*) and acoustic signals. The acoustic power
spectrum illustrates the complexity of the dynamics, as there are multiple local peaks in
77
the spectrum. For example, there are multiple modes (peaks) at approximately 580, 660
and 725 Hz, as well as their harmonics. The strongest mode at 660 Hz corresponds to the
axial quarter wave mode of the combustor, based on an assumed uniform gas temperature
of ~1700 K. It has a sound pressure level of nearly 120 dB. In the optical power
spectrum, the power is distributed over a similar broad range of frequencies, with a lower
fraction of the power at the instability frequencies. For more pronounced combustion
dynamics, nearly all the power would be expected to be in narrow ranges around the
instability frequencies, similar to the spectra in the gas-fueled combustor. While only
2 atm results are shown here, 4 atm operation produces a similar behavior.
4.2.2. Precursor events
1.6
Amplitude (a.u.)
1.4
1.2
0.8
0 10 20 30 40 50 60
Time (ms)
Figure 46. CH* signal time trace with precursor event features indicated by arrows
for combustor operation close to LBO at Φ/ΦLBO = 1.02.
Figure 46 displays a CH* time trace for combustor operation at 2 atm just prior to
LBO (at Φ/ΦLBO = 1.02) with precursor event-like features indicated by arrows. The
signal features are characterized by temporarily low amplitude, for an extended duration,
compared to the neighboring region; this is the expected behavior for LBO precursors.
However, the amplitude drop is not significantly lower than the nominal signal minima.
Thus the precursor signatures observed here are less pronounced than those observed in
the gas-fueled combustor. The duration of these precursors is about 2-3 ms, much smaller
78
than the ones in the gas fueled combustor (15-45 ms). Similar short duration precursors
(~6 ms) were observed in prior LBO sensing studies in a conventional (non-LDI) liquid
fueled combustor [25]. Besides, unlike the gas fueled combustor, signal amplitude stays
well above zero during precursors. The smaller durations and modulation depths of
precursors result in smaller precursor signatures. The smaller signatures require a careful
selection of the event detection algorithm to prevent false detection of events.
Precursors with much weaker signatures (smaller durations and modulation
depths) than the ones shown in Figure 46 were also observed, especially for conditions
not so close to LBO, e.g., Φ/ΦLBO=1.09. For providing LBO warning with a sufficient
margin, such events have to be detected, without producing false events far from LBO.
Reducing the “noise” (any amplitude modulation other than the precursor) can be
expected to minimize false events. The “noise” in the optical signal has contributions
from both dynamic instability and turbulent combustion generated noise. While the
turbulent noise is broadband, the instability modes have periods in the range 1.4-1.7 ms
(580-725 Hz). On the other hand, the precursor event durations are in the range 1.5-3 ms.
The short duration events have similar time scales as the dynamic instability period. Low-
pass filtering at 500 Hz, corresponding to a 2 ms time period, is able to suppress the
instability and other high frequency noise and improved precursor signatures. This is
evident in Figure 47, which presents a low-pass filtered optical signal corresponding to
the raw signal in Figure 46. Low-pass filtering makes the precursors more evident. The
filtering was performed with an 8th order digital Butterworth filter. The higher filter order
was used to achieve a sharper cutoff, as the instability period and precursor durations are
close to each other. The filtering delays precursors by 2 ms, which may not be significant.
79
1.2
Amplitude (a.u.) 1.1
0.9
0.8
0 10 20 30 40 50 60
Time (ms)
Figure 47. 500 Hz low-pass filtered CH* signal with precursors indicated by arrows
for combustor operation close to LBO at Φ/ΦLBO = 1.02.
High-speed flame images during such a precursor event are shown in Figure 48.
The images are successive 1 ms exposures, with the camera viewing at an angle of 45°
from downstream direction, similar to the optical fiber (see Chapter 3). Grayscale (0-255
range) images are scaled with an intensity based color map as shown in the figure. The
images are broadband chemiluminescence images, filtered between 325-600 nm using an
optical filter placed before the camera lens. The 4th and 5th frames show a signification
amount of flame extinction, followed by its recovery in the subsequent frame. Far from
LBO, no such extinctions were observed and the flame appearance is similar to that of the
flame in first three frames, with some unsteadiness. From observations of the high speed
images during other extinction events, there is no single, preferable region where
extinction occurs. In fact, it seems that the extinction event may begin at any particular
azimuthal location and rotate around the combustor during the event. Based on the high
speed images, the duration of the extinction is approximately 2-3 ms, as some level of
extinction may still be present in the 3rd and 6th frames as well. Unlike the gas fueled
combustor, there is no evidence of flame lift-off during precursors. Absence of such a
flame mode change may be the cause of shorter event durations for the LDI combustor.
80
Figure 48. High speed flame images during a precursor event in the LDI combustor
at 2 atm operation near LBO.
4.2.3. LBO Proximity Sensing
Precursor events in the filtered optical signals can be detected using the double
thresholding approach, similar to the one used for the gas-fueled combustor. However,
due to the smaller modulation depths and the durations of the precursor events, the
threshold level and the duration constraints may require modification. Results of event
identification for different lower threshold settings, obtained from low-pass filtered CH*
signals, averaged over 39 s at each equivalence ratio are shown in Figure 49. In the
figure, an inset plot with the vertical scale reduced is also shown in order to emphasize on
the event rates far from LBO. The upper threshold is kept at -0.5, and the duration
constraint at 1.4 ms. In evaluating the signal mean and standard deviation, any departure
above and below the -3 limit is excluded. Besides, the signal during precursor events
is also excluded from the calculation. From the figure, though all the threshold settings
seem to provide an acceptable event rate trend, the setting -3.5 may be the preferable
choice, as it produces nearly zero events far from LBO and maximum near LBO. The
setting -3.25 produces non-zero event rate far from LBO, i.e., in the Φ/ΦLBO range 1.3-
1.5, and hence may not be preferable. On the other hand, the -3.75 threshold setting,
81
though it produces zero events far from LBO, has a lower event rate near LBO and hence
is not preferable.
15
14
Average number of events/s  -3.25
12  -3.5
 -3.75
10 1
8 0.8
0.6
6
0.4
4 0.2
2 0
1 1.2 1.4 1.6
0
1 1.1 1.2 1.3 1.4 1.5 1.6
Φ/ ΦLBO
Figure 49. Average event rates for different lower threshold settings; upper
threshold=-0.5 and minimum time constraint =1.4 ms.
Besides the lower threshold, the event rate was also observed to be sensitive to the
duration constraint as the precursor durations are close to the noise time scales. Average
event rates for different duration constraints are plotted in Figure 50. The lower threshold
in all cases is -3.5 and the upper threshold is -0.5. Though all the duration
constraints produce an acceptable event rate trend, the 1.4 ms constraint produces nearly
zero events far from LBO and a high level near LBO. Based on these results, a lower
threshold of -3.5, an upper threshold of -0.5 and a minimum duration constraint of
1.4 ms was chosen to be the optimal settings for event detection in low-pass filtered CH*
signals in the LDI combustor.
82
15
14 1.2 ms

12 1.4 ms
1.6 ms
10
1
8 0.8
0.6
6
0.4
4 0.2
2 0
1 1.2 1.4 1.6
0
1 1.1 1.2 1.3 1.4 1.5 1.6
Φ/ ΦLBO
Figure 50. Average event rates for different minimum duration constraints; lower
threshold=-3.5 and upper threshold=-0.5.
12
2atm
10 4atm
0
1 1.1 1.2 1.3 1.4 1.5
Φ/ ΦLBO
Figure 51. Average event rates for 2 and 4 atm operations.
Average event rates for 2 and 4 atm operation, obtained from 500 Hz low-pass
filtered signals, are compared in Figure 51. For both pressures, the optimal event
detection settings, identified above are employed. The average event rate is nearly the
same for both the operating conditions. The seemingly lower event rate very close to
83
LBO for 4 atm operation is due to the lowest equivalence ratio not being as close to LBO
as for the 2 atm case. The similar event rates for the different operating pressures are an
indication of the robustness of this LBO sensing approach.
In the gas-fueled combustor, event durations and modulation depths increased
towards LBO. Similar results for the liquid-fueled LDI combustor are plotted in Figure
52. Event durations are obtained from the unfiltered signals. As before, the duration of an
event is defined as the time spent by the signal below the local mean during an event.
Events are detected with a lower threshold of -3.25, an upper threshold of  and a
duration constraint of 0.8 ms. These settings provided event rates similar to the optimal
event rates obtained from the filtered signals. The modulation depths are obtained from
low-pass filtered signals, using the optimal threshold settings for low-pass filtered
signals. Both event duration and modulation depth remain nearly constant until very close
to LBO, then increase. Therefore combining event duration and modulation depth with
event occurrence rate, using the Stability Index, should provide a more robust LBO
proximity parameter, similar to the gas-fueled combustor.
Stability Index and event rate, normalized with the corresponding average values
far from LBO (Φ/ ΦLBO =1.25-1.47), are plotted in Figure 53, along with absolute values
for SI. Both the event rate and SI are obtained from low-pass filtered signals, using the
optimal threshold settings. SI is calculated by integrating the signal loss below the local
mean during precursor events. As found for the gas-fueled combustor, SI provides a
higher dynamic range compared to the average event rate. However, the LDI SI is about
an order of magnitude smaller than the SI in the gas-fueled combustor. This is due
primarily to the shorter event durations and modulation depths in the LDI combustor.
84
2.5 0.4
Average duration of an event (ms)
Normalized average modulation

Modulation depth
Duration
0.3
2
depth
0.2
1.5
0.1
1
0
1 1.1 1.2 1.3 1.4 1.5 1.6
Φ/ ΦLBO
Figure 52. Average duration and normalized modulation depths of an event.

600 0.8
Normalized event rate and SI
Events
500 SI

0.6
400
300 0.4
200
0.2
100
0 0
1 1.1 1.2 1.3 1.4 1.5 1.6
Φ/ ΦLBO
Figure 53. Average event rate and Stability Index (SI) comparison. Left axis scale:
event rate and SI normalized with average values in Φ/ ΦLBO range: 1.25-1.47.
The potential for real time LBO proximity sensing during slow transients, using
precursor events, is demonstrated in Figure 54. Here, the fuel flow rate was decreased
slowly and continuously, starting from t=0 s until LBO occurred at t=63 s. The number of
events and SI at a time t are obtained from the low-pass filtered CH* signal, during t-1 s
to t. The same optimal event detection settings presented above are used here. Far from
85
LBO, the number of detected events is nearly zero, with occasional events seen between
Φ/ΦLBO=1.45-1.2. Events start to appear frequently at Φ/ΦLBO=1.2 (t=35 s). The event
occurrence rate appears to vary widely at this point, ranging from 1/sec to 5/sec between
Φ/ΦLBO=1.3 and 1.1 (i.e., t=60 to 68 sec). It then rises sharply at Φ/ΦLBO=1.1. As
expected, the SI provides a higher dynamic range, as events and SI are matched for
Φ/ΦLBO=1.1-1.06 (t=34-42 s). In addition, the SI appears to have a smoother, less noisy
trend near LBO, compared to event occurrence rate.
The result demonstrates that an optical sensor monitoring LBO precursor events
can be used as an input to a real-time LBO control system.
1.6
/
LBO
1.4
Φ/ ΦLBO
1.2
1
0 10 20 30 40 50 60

Number of events/sec
30 Event rate 2
SI
10 0.5
5
LBO 0.1
2
1
0.75 0.02
0 10 20 30 40 50 60
Time (s)
Figure 54. (top) Normalized equivalence ratio variation with time (bottom) Number
of events and SI calculated in moving 1 s time windows.
4.2.4. CH* vs OH* Signals
The above results were presented for event detection using CH*
chemiluminescence; the OH* signal is an alternative as both are generally proportional to
86
heat release rate. Precursor events detected in low-pass filtered CH* and OH* signals
using the same event detection settings are presented in Figure 55. Both optical signals
were acquired simultaneously with the same optical fiber probe, with the light exiting the
probe split and sent to two spectrally filtered detectors, one for CH* and the other for
OH*. With the same event identification settings, OH* produces slightly higher event
rates (0.3/s) far from LBO, whereas CH* produced nearly zero events. In addition, OH*
produces marginally higher event rates near LBO compared to CH*. With the threshold
settings modified to lower the OH* event rate far from LBO, the event rate near LBO is
reduced to to about ~7/sec. This suggests some differences in the background CH* and
OH* emissions, especially far from LBO. Since CH* signals provided a better LBO
proximity parameter in the present study, they are the focus of the results presented in
this thesis.
14
Normalized event rate and SI
CH*
12 OH*
10
0
1 1.1 1.2 1.3 1.4 1.5 1.6
Φ/ ΦLBO
Figure 55. Average event rate from low-pass filtered CH* and OH* signals with
same threshold settings.
Normalized unfiltered and low-pass filtered CH* and OH* signals, during a
precursor detected in the filtered OH* signal, far from LBO at Φ/ΦLBO=1.54, are shown in
Figure 56. In the filtered OH* signal, the amplitude drop indicated by the arrow was
detected as an event. However the corresponding filtered CH* signal does not show any
87
amplitude drop. Examining the unfiltered signals OH* signal shows amplitude drop
similar to precursors, whereas CH* does not show any such drop. The difference in signal
features for OH* and CH* is not surprising, as the two respond somewhat differently to
variations in heat release and unsteady combustion processes. For example, CH* and OH*
emission respond differently to perturbations in local equivalence ratio and strain [79].
1.2
0.8
(a.u.)
0 5 10 15 20 25 CH*30
OH*
Amplitude
1.1
0.9
0.8
0 5 10 15 20 25 30
Time (ms)
Figure 56. CH* and OH* signals during a precursor detected in the filtered OH*
signal; (top) unfiltered (bottom) low-pass filtered signals. (Φ/ ΦLBO=1.54).
4.3. Summary
This chapter investigated LBO margin sensing under high amplitude dynamic
instability and under nearly realistic combustor operating conditions, i.e., elevated
pressure and temperature operation. The LBO margin sensing approach was based on
identifying flame extinction and re-ignition events (LBO precursors) in
chemiluminescence signals.
LBO precursor events similar to those observed in dynamically stable conditions
were also observed under high amplitude dynamic instability conditions, with two
88
distinct instability mechanisms (w/o Φ' and w/ Φ') produced by well-mixed and closed-
coupled fueling in a gas-fueled combustor. Low-pass filtering applied to optical signals
produced similar characteristic shape for precursors for all conditions, allowing for a
single LBO precursor detection algorithm to be applied, as the results are relatively
insensitive to changing level or type of dynamics. Precursor events detected in optical
signals were observed to increase in number as LBO conditions were approached,
providing a measure of proximity to LBO. Another parameter, Stability Index (SI), which
combines event rate, duration and strength, produced a more robust LBO proximity
parameter having higher dynamic range than the simple precursor event occurrence rate.
On average, event rate near LBO was observed to be in the range of 1-2/sec and event
durations in the range 20-50 ms.
The same LBO precursor detection approach was applied to a different combustor
configuration, a liquid-fueled LDI combustor, operating at elevated pressure and preheat
temperature. The combustor was observed to have a moderate level of dynamic
instability, with multiple frequencies. Near LBO, the combustor exhibited partial
extinction and re-ignition events. These events were observed to have much shorter
durations (1.5-3ms) compared to the gas fueled combustor. The durations are similar to
dynamic instability time periods (1.4-1.7ms). In optical signals some of these events were
observed, however, others were obscured to some extent by dynamic instability and other
high frequency noise. Nevertheless, low-pass filtering enabled robust event detection. In
optical signals CH* produced more number of events near LBO compared to OH*,
suggesting CH* is a better choice. The detection approach was demonstrated to be
capable of detecting LBO margin in real-time.
89
CHAPTER 5
LBO MARGIN SENSING: ACOUSTIC DETECTION
In addition to optical emissions, acoustic emissions can be used for monitoring
LBO precursor events, as acoustic radiation is proportional to the time derivative of the
heat release rate. During precursor events, the drop and rise in heat release can be
expected to produce a corresponding acoustic signature. This chapter examines precursor
event detection and LBO margin sensing using acoustic signals in the gas-fueled and LDI
combustors. Acoustic data was acquired simultaneously with the optical results presented
in the previous chapter. In the gas-fueled combustor, acoustic signals were acquired with
a microphone located outside the combustor, while in the LDI combustor they were
acquired with a piezo-electric transducer mounted on a standoff tube as described in
Chapter 3.
5.1. LBO Margin Sensing in the Gas-Fueled Combustor
An acoustic signal and a simultaneous OH* chemiluminescence time trace during
a precursor event for dynamically stable operation are shown in Figure 57. The acoustic
data exhibit a low level of coherent oscillations, which are less evident in the optical
signal. The time delay between the acoustic and optical signals is about 2-3 ms based on
the acoustic travel time from the combustor to the microphone. In the acoustic signal, a
precursor event signature can be seen around the time when the optical signal shows re-
ignition. The extinction phase does not produce a clear acoustic signature, probably due
to the slow rate at which the heat release drops in this case. The re-ignition appears more
rapid, with the heat release rate overshooting the nominal heat release rate, probably due
to rapid burning of reactants present in the extinguished region. It should be noted that
the optical signal here corresponds to a small region above the combustor inlet, thus it
90
represents local heat release. On the other hand, the acoustic signal is due to global heat
release variations.
3
OH*
2
1
Amplitude (a.u.)
0
0 25 50 75 100 125 150
4 Acoustic
2
0
-2
0 25 50 75 100 125 150
Time (ms)
Figure 57. (Top) Precursor event in the OH* signal for dynamically stable operation
with a precursor event at ~50-90 ms; (bottom) simultaneous acoustic signal.
4
2
0
-2
-4
Amplitude (a.u.)
0 50 100 150 200 250
-5
0 50 100 150 200 250
Time (ms)
Figure 58. Acoustic signals during precursor events observed in optical signals
(Figure 32) for instability w/o Φ' and w/ Φ'.
91
Acoustic signals, acquired simultaneously with the optical signals of Figure
32(Chapter 4) for instability w/o Φ' and w/ Φ', are shown in Figure 58. The time windows
where optical precursors were observed are indicated by boxes. For both instability cases,
the oscillation amplitude is reduced during the precursor. However, the reduction in
oscillation amplitude is not limited to the precursors. For example for instability w/o Φ',
the oscillation amplitude drops around 75 ms and 230 ms, though the optical signal did
not show any corresponding precursors. In addition, the oscillation amplitude drops
frequently for instability w/ Φ' due to its inherent amplitude modulation. Therefore, a
reduction in dynamic instability amplitude cannot be considered a precursor event
signature, as it is not uniquely attributable to the occurrence of precursor events. From a
visual inspection, there is also no other clear sign of a precursor event in the raw acoustic
signals during strong combustion dynamics.
Given the success of low-pass filtering for suppressing the combustion dynamic
component in the optical signals, low-pass filtering at 50 Hz was applied to the acoustic
signals for the dynamically unstable cases. The results, along with a result for a
dynamically stable case, are shown in Figure 59. These traces correspond to the raw
acoustic signals in Figure 57 and Figure 58. Low-pass filtering was performed with an 8th
order digital Butterworth filter. In the figure, the signal amplitude of the stable mode was
reduced by a factor of 5 to put it on the same scale with the unstable cases.
Though there was no indication of precursor events in the raw acoustic signals
during instability, clear precursor event signatures can be seen in the filtered signals. It is
likely that in the raw acoustic signals, precursor events have much smaller amplitudes
than the instability and are therefore obscured. By suppressing the instability, the
precursor signature is revealed. All three precursors have nearly the same characteristic
shape, and hence the same event detection algorithm can be used for their detection. For
the precursors, the downward portion of the event signature is more pronounced than the
92
upward portion, possibly due to the effect of filtering. Since the upward part of the
signature comes later in time, filtering would tend to add a portion of the downward
signal to the upward, thus suppressing the latter more. For instability w/ Φ', the upward
modulation is relatively smaller than the other two cases, and this behavior has been
observed to be consistent for many other precursors. The smaller upward modulation
could be due to a faster time scale for re-ignition, and thus it is suppressed more by low-
pass filtering.
0.05
Amplitude (a.u.)
-0.05 No instability
Instability w/o '
-0.1 Instability w/ '
0 50 100 150 200 250
Time (ms)
Figure 59. Low-pass filtered acoustic signal with precursor events for dynamically
stable and unstable w/o Φ' and w/ Φ'.
Precursor events in the filtered acoustic signals can be detected based on the
downward modulation feature alone, as the upward modulation is not as strong. Results
of event identification for all three cases of operation, obtained from low-pass filtered
signals, are shown in Figure 60. For event identification, a lower threshold of -4 was
used. In the threshold calculation, amplitude excursions above and below the 3 limit and
the signal during precursors are omitted.
93
Dynamically stable

1.5
Instability W/O 
Instability W/ 
0.5
0
0 0.02 0.04 0.06 0.08 0.1
Figure 60. Average event rates obtained from low-pass filtered acoustic signals for
dynamically stable and unstable conditions.
Events for the dynamically stable mode and unstable mode w/o Φ' provide the
desired trend, a nearly monotonic increase in average event occurrence rate as LBO is
approached. In addition, event rates are similar to the rates obtained with optical
detection. On the other hand, for instability w/ Φ', the event rate increases initially
towards LBO, but then drops. This is due to the lack of a significant precursor signature
for very long duration precursors, which occur close to LBO. The long duration
precursors did not produce noticeable precursor signatures, probably due to the slower
changes in heat release compared to the other two cases.
5.2. LBO Margin Sensing in the LDI Combustor
For comparison, an acoustic signal from the LDI combustor is shown in Figure
61, along with its 500 Hz low-pass filtered version. Results are shown during times when
a precursor event was observed in the optical signal (Figure 46-Chapter 4). The time
delay between the `optical and acoustic signals is about ~1.5 ms, based on the travel time
to the acoustic pressure sensor. The time windows where the precursor signatures are
94
expected (from the optical detection) are indicated by boxes. The behavior of the raw
acoustic signal during these instances is not noticeably different from other times. The
filtered signal does not show significant improvement either, except for slightly longer
features. Unlike the gas-fueled combustor results, where filtering made precursor
signature conspicuous, the same behavior was not observed for the LDI combustor.
The absence of clear precursor signatures in the acoustic signal for the LDI
combustor could be due to smaller changes in heat release during precursors. Compared
to the gas-fueled combustor, the extent of flame extinction and consequently the re-
ignition is much smaller. This would produce smaller fractional variations in heat release,
and therefore weaker acoustic signatures. Another possible reason is that in the LDI
combustor, the dynamic instability period is comparable to the precursor event duration,
which would limit the effectiveness of filtering in separating precursors from dynamics.
Finally, reflections of the precursor pulse from combustor boundaries may play a role as
well. These latter two issues are examined in the following sections.
Un-filtered
0.2
0
Amplitude (a.u.)
-0.2
0 10 20 30 40 50 60
Low-pass filtered
0
-0.1
-0.2
0 10 20 30 40 50 60
Time (ms)
Figure 61. (Top) Raw acoustic signal time trace from the LDI combustor during
precursor events observed in optical signals, as marked by boxes; (bottom) low-pass
filtered signal.
95
5.3. Effect of Instability and Precursors Timescales
When the duration of precursor events and the dynamic instability period are
sufficiently different, filtering can be applied to separate precursors from dynamics.
However as the time scales of both get closer, filtering can be expected to be ineffective.
This is one reason filtering may have failed to reveal precursor signatures in the LDI
acoustic signals. The problem might be expected to exist for optical signals as well, if
precursors have weak signatures and the dynamics are strong enough to obscure them. To
address possible differences between the two detection approaches, we can analyze the
limitations of filtering when the instability period and precursor event durations are
nearly the same. Since filtering separates signals based on their spectral content, the
spectra of precursors relative to that of the dynamic instability needs can be analyzed.
Ideally, combustion dynamics produces harmonic oscillations in heat release and
pressure. The signals usually have spectra that are narrowband, with peaks at the
instability frequencies. However, amplitude variations and phase drifts can result in some
broadening of the instability spectra. Precursor events are discrete and occur
intermittently at a low rate (usually below 10-20 events/sec). Such discrete events have
wide frequency content due to their non-periodic nature [80]. This behavior is similar to
spectral leakage arising from using a finite time interval for spectral evaluation of a
periodic signal. For example, if spectrum of a periodic signal is evaluated over only one
cycle, the spectrum would be broad due to convolution with the spectrum of a boxcar
truncation function [80].
Model optical and acoustic precursor event time traces, along with their amplitude
spectra are plotted in Figure 62. An optical precursor event is generally an inverted,
single peak curve, and can be modeled by Eq. (5.1). The acoustic precursor source event,
given by the time derivative of the optical precursor, is given by Eq. (5.2). In the figure,
the duration of precursor events is normalized to have a unit time scale. Similarly in the
96
spectra, a unit frequency corresponds to the inverse of the precursor time scale. As seen
in the amplitude spectra, both precursors have quite broad spectra. For a dynamic
instability having the same period as the precursor duration, the spectrum would be
narrowband with unit center frequency. This suggests that even if the durations of the
precursors and the instability period are similar, filtering (low-pass or a narrow bandpass
around instability frequency) would still be effective in improving precursor signatures.
The acoustic precursor has a relatively higher amount of power above unit frequency
compared to the optical precursor. Hence, low-pass filtering will tend to suppress
acoustic precursors more than the optical precursors.
Q  t   1  0.5e
 t    2.5  t  2.5
2
(5.1)
 t 
p  t   te
2
   t  2.5 (5.2)
-3
x 10
1 1
0.8 0.5
Amplitude spectral density (a.u.)
0.6
0
-0.5 0.5 0 1 3
Amplitude (a.u.)
(b)
(a)
-3
x 10
0.5 1
0 0.5
-0.5 0
-0.5 0.5 0 1 3
(c) (d)
Normalized time Normalized frequency
Figure 62. Precursor events and their spectral content: (a) optical precursor, (b) its
amplitude spectrum, (c) acoustic precursor, and (d) its amplitude spectrum.
97
The effectiveness of filtering in revealing precursors in the presence of
instabilities, when both have the same time scales, is illustrated in Figure 63. The
instability signal is modeled by a sine function with a unit period, having a peak-to-peak
amplitude of 2 units. In addition, Gaussian noise having a standard deviation of 0.1 units
is added to the instability signals. The downward peak of the optical precursor has an
amplitude of 0.5 units, whereas the acoustic precursor has a peak-to-peak amplitude of 1
unit. Optical and acoustic precursors are added to the instability signal, at a time of 5
units, which corresponds to the zero phase of the instability signal. Raw signals, low-pass
filtered with an 8th order Butterworth filter at a cutoff of frequency of 0.6 units
(instability frequency =1 unit), are also shown in the figure.
3
1
2
0
1
-1
0
Amplitude (a.u.)
Amplitude (a.u.)
0 5 10 15 0 5 (b) 10 15
(a)
1.6 0.1
1.4 0
-0.1
0 5 10 15 0 5 10 15
(c) (d)
Time (a.u.) Time (a.u.)
Figure 63. Model optical and acoustic signals with instability and precursors having
similar time scales: (a) optical instability signal with a precursor; (b) acoustic
instability signal with a precursor; (c) low-pass filtered optical signal; and (d) low-
pass filtered acoustic signal.
98
In the raw signals, there is no discernible precursor. However, the filtered signals
clearly show the precursors. This is due to the above observation that precursors have
wider spectral bands compared to the instability. In the filtered results, the signal-to-noise
ratio is higher for the optical precursor compared to the acoustic precursor. This is due to
the acoustic precursor having more frequency content above the filter cutoff of compared
to the optical precursor. Thus the acoustic precursor is suppressed more by the low-pass
filtering. In addition, similar simulations show that the shape of the filtered optical
precursor is relatively insensitive to the phase at which the precursor occurs. On the other
hand, the relative magnitudes of the downward peak compared to the upward peak of the
filtered acoustic precursor are sensitive to the phase.
5.4. Effect of Reflections on Acoustic Precursor Detection
The acoustic pressure measured by a transducer at any given time is not due
solely to acoustic emission produced at a single instance. It has components of acoustic
emissions from a range of previous times, due to reflections from boundaries. This is less
of an issue for optical detection, because the propagation time of light makes any
reflections nearly simultaneous with the original. Thus one might expect that in addition
to the acoustic precursor pulse traveling directly from the source, i.e., the location in the
flame where extinction and re-ignition occur, there will also be reflections of the acoustic
pulse from the combustor boundaries. These reflections can overlap with the actual
precursor signal at the transducer and may corrupt the expected precursor signature.
5.4.1. Acoustic Model
To examine the effect of reflections on detection of acoustic precursors, a simple
combustor model was developed (see Figure 64). The model consists of two regions with
different impedances, Z1 and Z2, separated by a thin flame. The flame is assumed to stand
at a distance of L1 from the inlet. The two impedances, Z1 and Z2, are given
99
by Z  C where ρ is density of the medium and C is the speed of sound. The impedance
to the right of the tube exit boundary is given by (5.3), which takes into account the end
correction and mean flow for an un-flanged tube [81]. In that equation, k is the wave
number, a is the tube diameter and M is the flow Mach number. Though the precursor
pulse has a range of wave numbers (frequencies), a single wave number is used, based on
the inverse of the precursor duration. Correction to the tube length due to the end wall is
also applied, as given by l  0.613a . Reflection and transmission coefficients for a
pulse traveling from left to right at a boundary are given by Eq. (5.4) and Eq. (5.5). The
reflection coefficient at the inlet boundary is calculated based on area change at the inlet
given by Eq. (5.6).
Figure 64. Schematic of acoustic model system used to investigate the influence of
reflections on the detected acoustic pressure signal. The flow inlet is at the left; two
possible transducer locations are indicated.
  (ka)2  
Z L  C    i0.613ka   M L  (5.3)
 4  
Z r  Zl
Rl ,r  (5.4)
Z r  Zl
Tl ,r  1  Rl ,r (5.5)
100
S1  S0
R1,0  (5.6)
S1  S0
It is assumed that the acoustic precursor originates at the flame and produces
waves traveling to the left and right. The left going wave gets partially reflected from the
inlet, while the right going wave is reflected from the exhaust. Also, each time a wave
goes through the flame, it becomes two waves, one reflected and the other transmitted.
This gives rise to numerous waves crisscrossing inside the combustor. Pulses are
dissipated at the inlet and tube exit due to the reflection coefficients being less than one.
Dissipation at the tube wall boundary layer by thermal and viscous losses is neglected
here, as it is usually small for short tube lengths [81]. The signal measured by a
transducer is a superposition of the multiple pulses separated by different delay times.
To simplify the implementation of this model, the relative strength of the
reflections can be considered. Reflections by the inlet and tube exit are considerably
stronger (R1,0 and R2,L ~0.9) than the reflections at the flame boundary (R1,2 ~0.25 for
LDI and ~0.4 for the gas-fueled combustor). Therefore up to 15 reflections by the inlet
and tube exit are accounted for in the current analysis, whereas flame reflections are
truncated after two or three reflections. For example in the LDI combustor model, the
amplitude has dropped to ~6% of the original pulse amplitude by the second flame
reflection, and by the third reflection, the amplitude is as low as 1%. Thus only two
flame reflections are considered. In the gas-fueled combustor, the 4th reflection amplitude
drops to ~3% and hence three flame reflections are included. The pressure transducer is
assumed to be mounted either at the inlet similar to the LDI setup or at the tube exit,
similar to the gas fueled combustor setup.
101
5.4.2. Model Results for LDI Combustor
The model parameters required to simulate the LDI combustor are: L1 = 30 mm,
L2= 270 mm, T1=700K, T2=1970K and P=2atm. The resulting reflection and transmission
coefficients are R1,0=0.89, R2,L= -0.88, R1,2= -0.25, R2,1=0.25, T1,2=0.75 and T2,1=1.25. In
calculating R1,0 , S0 included the open area of the swirler and the perforated plate hole
area. The acoustic precursor is assumed to have a shape as shown in Figure 62(c), with a
duration of 2 ms, and it is created at time zero.
The original precursor generated by the flame, i.e., reaching the transducer
without any reflections and denoted the initial precursor, and the precursor signal with
reflections are shown in Figure 65. Reflections produce a ringing behavior in the
precursor signal seen by the transducer. Ringing is due to the reflected precursor pulses
reaching the transducer with different time delays and the decay of the amplitude by each
reflection. Thus the ringing is periodic with a slow decay, and resembles a typical
instability signal. The amplitude spectra of the initial and reflected precursors are shown
in Figure 66. Unlike the initial precursor with its power distributed over a broad range of
frequencies, the reflected precursor has the majority of its power in a narrow band,
centered at ~610Hz. In addition at low frequencies (<300 Hz), the reflected precursors
has nearly half the power of the initial precursor. The natural frequency of the combustor
duct can be calculated using the boundary conditions P'1= P'2 and ρ1U1'= ρ2U2' at the
flame boundary along with U'0=0 at the inlet and P'L=0 at the tube exit. The calculated
fundamental natural frequency of the combustor duct is 630 Hz. Therefore reflections
result in a precursor signal having nearly the ringing frequency as the combustor’s natural
frequency, which is usually the frequency for combustors exhibiting longitudinal
dynamic instability. This makes it hard to separate the precursor from dynamic
instabilities. This result, reflections generating a precursor signal at the duct natural
frequency, is not surprising. For example, for a closed open duct, the periodicity of pulses
reflecting back and forth is 4L/C and the inverse of it is the natural period of the duct.
102
1
Reflected
Amplitude (a.u.)
0.5 Initial
Single
-0.5
-1
0 5 10 15
Time (ms)
Figure 65. An acoustic precursor reaching the transducer without any reflections
(initial precursor) and the resultant of multiple reflections (reflected precursor).
-3
x 10
6
Initial
Single
5
Reflected
A.S.D. (a.u.)
0
0 500
1000 1500
Frequency (Hz)
Figure 66. Amplitude spectra of the initial and reflected precursors.
In order to investigate the limitations of filtering for separating the reflected
precursor from dynamics, the reflected precursor is added to an instability signal at 630
Hz. The peak-to-peak amplitude of the instability signal is taken to be 6 units. In addition,
Gaussian noise with a standard deviation of 0.05 units is added to the instability. Since
precursors cannot be identified in raw signals under instability conditions, low-pass
filtering was performed at 350 Hz with an 8th order Butterworth filter. Raw signals along
with low-pass filtered signals are shown in
103
Figure 67, for initial and reflected precursors added to the instability signals. The
precursors added to the instability have amplitudes as shown in Figure 65. In the filtered
signals, the initial precursor simulation results in a more distinct signature compared to
the reflected precursor (same vertical axis scale for both). Though the reflected precursor
is somewhat clear in this simulation, it may not be as evident in real combustor scenarios
where combustion dynamics are more complex, with instability frequencies having a
broader spectrum than simulated here. In addition, this analysis assumed plane wave
propagation. In reality due to the non-planar nature of the initial precursor, reflections
from the combustor duct walls may contribute to increased noise in the reflected
precursor signal.
5
0.05
0
0
-0.05
-5
Amplitude (a.u.)
Amplitude (a.u.)
0 10 20 30 40 0 10 20 30 40
(a) (c)
5
0.05
0
0
-0.05
-5
0 10 20 30 40 0 10 20 30 40
(b) (d)
Time (a.u.)
Time (a.u.)
Figure 67. Model instability signals with initial and reflected precursors added to
them: (a) instability signal with initial precursor; (b) instability signal with reflected
precursor; (c) low-pass filtered version of (a); (d) low-pass filtered version of (b).
The signal characteristics of the reflected precursors depend on the duration of the
precursor and the travel time inside the combustor, which depends on the combustor
length. For example, Figure 68 shows reflected precursor signals with the initial precursor
having different durations, in the same combustor geometry. Long duration precursors
104
(e.g., 8 ms) do not produce a noticeable ringing signal. This is due to the reflections
having much shorter time scales than the precursor duration. All the reflections, before
dying out, arrive nearly coincident with the initial precursor (this is closer to what would
be expected for optical detection). The reflected precursor has a somewhat lower
amplitude than the initial precursor. However, the precursor could be detected as it does
not coincide with the instability signal. If the tube was long enough, the 8 ms precursor
would also produce ringing, as the travel time becomes comparable to the precursor
duration.
1
2ms
0.5 5ms
Amplitude (a.u.)
8ms
0
-0.5
-1
0 5 10 15
Time (ms)
Figure 68. Simulated reflected precursor signals with the initial precursor having
different durations in the LDI combustor.
5.4.3. Model Results for Gas-Fueled Combustor
The model parameters that match the conditions of the gas-fueled combustor are:
L1 = 50 mm, L2= 550 mm, T1=300 K, T2=1830 K and P=1 atm. The resulting reflection
and transmission coefficients are: R1,0 = 0.97, R2,L= -0.92, R1,2= -0.42, R2,1=0.42,
T1,2=0.57 and T2,1=1.42.
Reflected precursors for different initial precursor durations are plotted in Figure
69. Long durations, e.g., above 10 ms, do not produce noticeable ringing. As such, we
can conclude that the actual precursor durations in the gas-fueled combustor (generally
above 5 ms) can be considered relatively large. Therefore, low-pass filtering helped with
105
precursor event identification in the gas-fueled combustor by separating the precursor
from the dynamics.
1 5ms
12ms
8ms
Amplitude (a.u.)
0.5
8ms
12ms
0
-0.5
-1
0 5 10 15 20 25 30 35
Time (ms)
Figure 69. Simulated reflected precursor signals with the initial precursors having
different durations in the gas-fueled combustor.
In addition to the above observations, the location of the pressure transducer
along the combustor wall is an additional concern. In the gas-fueled combustor, the
pressure transducer was located outside the combustor, which is qualitatively equivalent
to locating it at the tube exit. On the other hand in the LDI combustor, the pressure
transducer was located at the inlet. Both combustor geometries are approximately closed-
open tubes, which produce an acoustic anti-node at the closed end (inlet) and a node at
the open end (exit) for the standing wave expected for the case of combustion dynamics.
The precursor pulse amplitude, however, is independent of axial location. Thus the higher
pressure produced by the combustion dynamics at the inlet can more easily corrupt the
precursor signals for the case of the pressure transducer at the inlet compared to the case
where the transducer is near the exit. Hence for better detection of acoustic precursor
events, the pressure transducer should be located near the combustor exit, where pressure
oscillations associated with combustion dynamics are less pronounced.
106
5.5. Summary
LBO margin sensing using acoustic signals was investigated in the gas-fueled and
LDI combustors. In the gas-fueled combustor, in the presence of strong combustion
dynamics, no visible precursor signature was observed in the raw acoustic signals. Low-
pass filtering, however, provided clear precursor signatures by suppressing the dynamic
instability signal component. Precursors could be detected with a simple thresholding
approach based on the depth of (downward) signal modulation during events. In the LDI
combustor, neither raw signals nor low-pass filtered signals revealed reliable precursor
signatures. One possible reason is the smaller heat release changes during precursors in
the LDI combustor. Other possible reasons include, similar time scales for combustion
dynamics and precursors in the LDI combustor, reflection of precursors from impedance
discontinuities and, pressure transducer location.
Even if precursors have similar time scales as the combustion dynamic
oscillations, they can be separated due to the wider spectral bandwidth of the discrete
precursor events. The effect of precursor reflections from impedance discontinuities
inside the combustor were examined by modeling acoustic reflections in a one-
dimensional combustor geometry with an impedance jump caused by the flame and
combustor boundaries. The reflections result in a narrow band ringing signal with a
frequency similar to the combustor's natural frequency. This manifestation makes it
difficult to separate precursors from combustion dynamics. The ringing is not significant
if precursor durations are long compared to the round trip (acoustic) travel time inside a
combustor. Finally it is postulated that for better detection of precursors in the presence
of combustion dynamics, the pressure transducer should be located near the combustor
exit, where the combustion dynamic pressure oscillations have lower amplitudes.
107
CHAPTER 6
LBO MARGIN SENSING: RAPID TRANSIENTS
The previous two chapters investigated the existence and sensing of LBO
precursors in mostly steady-state operating conditions. In gas turbines, LBO could occur
both during nominally steady-state operation and rapid power reduction transients. The
current study, as well as previous efforts, have shown LBO margin can be effectively
sensed during slow variations in equivalence ratio. The same may not be possible for
rapid transients. Precursor events are discrete and occur intermittently. For a sufficiently
fast transient, for example if the fuel flow rate is rapidly reduced to lower the engine
power, a precursor may not occur before blowout occurs. Thus an approach is required to
analyze the likelihood (probability) of events occurring before blowout during a transient.
This chapter examines the appropriate probability models that can capture
statistical properties of precursor events. Using the statistical properties, probabilities for
events to occur during rapid transients is examined. In addition implementation of LBO
margin sensing during rapid deceleration transients in aircraft engines is examined.
6.1. Stochastic Model for Precursor Event Occurrence
At a given steady-state operating condition, precursor events occur sporadically,
with a wide variation in time between successive events. For example in the LDI
combustor at a nominally steady operating condition, the time between successive events
was as small as 10 ms to as high as 500 ms. Such a large variation in the time between
events is another indication of the random nature of the event occurrence, which should
be predictable with probability models. Since precursors are discrete events, a common
discrete event probability model such as Poisson distribution may predict the probability
108
for precursor event occurrence. The Poisson process has the property that the time
between successive events follows an exponential distribution.
Empirical cumulative distribution functions (CDF) of the time between
successive events (TBE), obtained from the LDI combustor data, are plotted in Figure 70.
Events were detected in the low-pass filtered optical signals with the optimized threshold
settings (lower threshold of µ-3.5 and a minimum duration constraint of 1.4 ms, see
Chapter 4). TBE is defined as the time between the starting times of successive events,
with the start of an event considered to be the time when the signal amplitude goes below
µ-0.25, during an event. In the figure, the empirical CDFs are compared with
corresponding exponential CDFs. The CDF of an exponential distribution is given by
Eq. (6.1), where λ is the average event occurrence rate (s-1), and τ is the time between
events (s). The event rate parameter λ is the inverse of the average time between events,
which was shown to be a function of the normalized equivalence ratio ( = Φ/ΦLBO) in
Chapter 4.
F ( )  P(TBE   )  1  e ( ) (6.1)
The empirical CDFs for equivalence ratios closer to LBO (Φ/ΦLBO =1.017 and
1.039) match quite well with the exponential CDFs, whereas for the equivalence ratios
farther from LBO, the agreement is not as good. However, the mismatch is simply an
artifact of having less data points (events) far from LBO. Sampled data probability
distribution functions qualitatively appear close to theoretical distributions only if there
are a sufficient number of data points to completely sample the distribution. Data for
equivalence ratios even farther from LBO are not compared, as the number of captured
events is even fewer, and a qualitative comparison could not be made. Though there is a
good agreement between the empirical and theoretical CDFs near LBO for most of the
time range, the theoretical distributions at smaller time scales, e.g., below 10 ms, have a
higher probability than the empirical results. This is primarily due to the precursors
109
having finite durations; therefore the time between successive events cannot be smaller
than this duration. There is no such constraint on the simple theoretical distribution used
here.
1
Φ/ΦLBO
0.8 1.094
1.084
Distribution function
+ 1.039
0.6
x 1.017
0.4
0.2
0
0.01 0.1 1 10
x
Time between events (s)
Figure 70. CDFs of measured time between successive events (TBE) for the LDI
combustor (symbols) along with corresponding exponential CDFs (lines).
Besides the qualitative comparison of CDFs, quantitative approaches such as
statistical hypothesis testing methods can be used for more accurate determination of the
distribution assumption. One such approach is the Kolmogorov-Smirnov hypothesis test
[82], which tests sampled data against its assumed distribution. The test is given by
Eq.(6.2), where F is the empirical distribution, F0 is the theoretical distribution, N is the
number of samples and α is a given significance level. The hypothesis assumption is
accepted if the maximum error between theoretical and empirical distributions is less than
the test metric given on the right side of Eq. (6.2). The test results for the data of Figure
70 are given in Table 3, where the commonly employed 5% significance level was used
for the test metric. The maximum error is 2.5-4 times less than the test metric for these
cases. The results further validate the use of an exponential distribution hypothesis.
110
1  
max F ( )  F0 ( )  log   (6.2)
2N 2
Table 3. Kolmogorov-Smirnov hypothesis test results.

Φ/ΦLBO 1.094 1.084 1.039 1.017
max|F(t)-F0(t)| 0.158 0.132 0.03 0.021
Test metric 0.409 0.32 0.134 0.067
1
Φ/ ΦLBO
0.8 1.60
1.47
1.35
0.6 1.20
1.09
1.04
0.4
1.02
Exponential
0.2
0
0.001 .01 0.1 1 10
Normalize time between events
Figure 71. Empirical CDFs of the normalized time between events, with events
detected using a threshold of µ-3. An exponential CDF with a unit mean time
between events is also shown.
Empirical CDFs of the time between events, with events detected using a lower
threshold closer to the signal mean (µ-3) and a duration constraint of 1 ms are plotted in
Figure 71. With the relaxed detection threshold, the number of detected events increases
significantly, with about 0.5-1 events/s events far from LBO and about 30 events/s near
LBO. However, it should be noticed that these results contain some false events as well.
In the figure, the time between events is normalized with the mean time between events
for each operating equivalence ratio. Normalized in this way, all of the distributions
111
collapse to a single curve. The measured data is compared to a normalized exponential
CDF having a unit mean time between events. With the modified thresholds and
increased number of events, useful for the statistical comparison, the empirical CDFs are
again in good agreement with the exponential CDF. The data again satisfy the
Kolmogorov-Smirnov hypothesis test. This indicates that the form of the probability
distribution does not change with small modifications of the threshold settings used for
event detection.
Empirical and corresponding exponential CDFs for the data from the gas-fueled
combustor are shown in Figure 72, for the case of instability without equivalence ratio
oscillations. Events were identified in the low-pass filtered optical signals with the
threshold settings that were observed to produce optimal event occurrence rate trend (see
Figure 41 Chapter 4). Similar to the LDI combustor results, the time between events is
well modeled by an exponential distribution. However very close to LBO
(Φ/ΦLBO=1.014), the empirical CDF appears to deviate from the exponential result for
smaller values of TBE (t<0.2 s). Such behavior is probably due to the increased fraction
of long duration precursors, as the precursor durations in the gas-fueled combustor are
much longer than in the LDI combustor. For the gas-fueled combustion, the hypothesis
testing results (not shown) again validate the use of an exponential distribution.
So for both the combustors, the time between events can be assumed with
reasonable accuracy to have an exponential distribution. As mentioned earlier, a process
characterized by discrete events, occurring randomly and with a TBE distribution
described by an exponential is known as a Poisson process. Therefore the stochastic
nature of event occurrence can be considered to follow Poisson statistics. Based on
Poisson statistics, the probability of the number of events being n during a time
interval t is given by the probability distribution function (PDF) in Eq. (6.3), where λ is
the average event occurrence rate. The Poisson probability distribution is a widely used
model for point processes, such as our precursor events. The model requires that events
112
occur independently of each other, which may be a reasonable assumption for the LBO
precursors, at least when events are not too close together.
e t  t 
n
P( N ev (t )  n)  (6.3)
n!
1
Φ/ ΦLBO
0.8 1.035
1.035
1.028
0.6 1.028
1.021
1.021
0.4
1.014
1.014
0.2
0
0.1 1 10
Time between events (s)
Figure 72. Empirical CDFs of time between events for the gas fueled combustor,
having instability without equivalence ratio oscillations, along with corresponding
exponential CDFs.
6.2. Probability of Precursor Event Occurrence in a Rapid Transient
The Poisson probability model described above is valid when the equivalence
ratio is kept constant, i.e., for a stationary Poisson process. During a power reduction
transient, the equivalence ratio changes with time. Assuming that at each point in the
transient, i.e., for a given Φ/ΦLBO, the stationary Poisson process would govern the event
rate, then the transient process can be described as a non-stationary or non-homogenous
Poisson process. For a transient having an equivalence ratio that changes with time, the
total number of events expected during the transient is given by Eq. (6.4), where Λ is the
113
expected number of events, T is the transient duration and  is the normalized
equivalence ratio (Φ/ΦLBO). In addition, the total number of events during the transient
follows the Poisson distribution given by Eq. (6.5), where N is the total number of events
in the transient. By specifying a given transient, (t), and for a given combustor’s (),
one can determine the probability of some minimum number (e.g., one or two) events
occurring in a given transient.
T
(T )     (t ) dt (6.4)
0
 n e
P  N (0  T )  n   (6.5)
n!
As an example, we can calculate the probability of events occurring within some
time during a transient in equivalence ratio for the LDI combustor. In this example, 
(Φ/ΦLBO) is reduced linearly from 1.20 to 1 at different rates. A value of =1.20 was
chosen for the starting condition because the event rate in the LDI combustor began to
rise at this equivalence ratio. The probability for at least one or two events to occur as a
function of duration of the transient is calculated using the non-stationary Poisson
distribution equations. A continuous function for event occurrence rate =fn(), required
for the calculation, is obtained by curve-fitting the steady state experimental data. In the
curve fit, the event rate far away from LBO is very small (<0.07 s-1), therefore the
contribution of false events to the calculated probabilities is negligible.
The probability of at least one or two events occurring is shown in Figure 73 as a
function of the transient duration. The results are presented for events detected with two
lower threshold values: µ-3.25 and µ-3.5. With a required probability of 98% for at
least one event to occur before LBO, the transient can have a duration no less than 2.35 s
for the µ-3.25 threshold case, and 3.35 s for the µ-3.5 threshold case. For an increased
114
requirement of at least two events occurring, with two events providing a more certain
indication of LBO proximity, the transients would have to be even slower (3.0 and 4.3 s
for the two threshold cases). Since the result for the minimum required transient duration
is contingent upon the value of  where the transient starts, a better descriptor might be
the equivalence ratio transient rate d/dt. For at least one event to occur, the maximum
allowed transient rate for the µ-3.25 threshold is 0.085 s-1 and 0.059 s-1 for the µ-3.5
threshold.
The probability for at least one or two events occurring in the gas-fueled
combustor are shown in Figure 74, again as a function of the duration of the power
transient. The transient starts at =1.06, where the event rate starts to increase. The
results shown are for the case of instability w/o Φ' and detection using the optimal
threshold settings (Chapter 4). With a 98% probability of at least one event occurring
before the end of the transient, i.e., before LBO occurs, the reduction must occur over at
least 4.85 s. The corresponding maximum transient rate is 0.012 s-1, much lower than for
the LDI combustor transient rate due to the lower event rate in the gas-fueled combustor.
115
1
0.98
Probability 0.95
0.9 3.5 , 1 Event

3.5 , 2 Events
3.25, 1 Event
3.25, 2 Events
0.85
1 1.5 2 2.35 3 3.35 4 4.5 5
Time (s)
Figure 73. Probability of at least one or two events occurring for a linear transient
in  from 1.20 to 1, as a function of the transient duration. Results are based on the
LDI combustor event rates.
1
0.98
Probability
0.95
0.9 1 Event
2 Events
0.85
4.85 2 6 37 4 8 9
Time (s)
Figure 74. Probability of at least one or two events occurring for a linear transient
in  from 1.06 to 1, as a function of the transient duration. Results are based on the
gas-fueled combustor event rates.
116
6.3. Precursor Event Simulation in Real-time
The characterization of the event rate as a Poisson process allows for simulations
to test various aspects of LBO precursor sensing. One such example is the simulation of
precursor events in real-time during a transient. This can be used to enable active LBO
control simulations, without having to do costly experiments. A few algorithms exist for
the simulation of non-stationary Poisson processes. However, only the algorithms capable
of real-time event simulations are useful for control simulations. A common approach for
non-stationary Poisson process event simulation is to use the thinning algorithm [83].
The algorithm requires the input of a maximum event rate (max) that can be expected in
the simulation. Next, the arrival time of the next event is generated using the maximum
event rate. The generated arrival time is accepted with a probability of (t)/max. If (t) is
small, such as what would be expected far from LBO, the generated event has a low
probability of acceptance. The simulation algorithm is given in Table 4.
The simulation approach is tested for its capability to simulate events for a slow
transient in equivalence ratio in the LDI combustor. The simulation results are presented
in Figure 75. In the simulation, () is obtained from a curve fit of the nominal steady-
state data. From ((t)) and the maximum event rate near LBO, the events are simulated
in real-time. The simulation results agree reasonably well with the experimental data. It
captures the general trend of fluctuations in number of events per second. It should be
noticed that the simulation result varies from run to run due to the random nature of the
simulation. The simulation slightly under- predicts the number of events farther from
LBO, and slightly over predicts the events very close to LBO. The difference may be due
to limitations of the simulation algorithm, and other simulation algorithms need to be
investigated.
117
Table 4. Event simulation algorithm for non-stationary Poisson process.
1) Start at t=0
2) Generate uniform pseudo random number U1[0,1]
3) Generate next event arrival time with the maximum event
rate with exponential distribution and reset t
1
to t  t  log U1 
max
4) Generate a second uniform pseudo random number
U2[0,1] independent of U1
 (t )
5) If U 2  then accept t
max
6) Go back to step 1
1.5 /
LBO
Φ/ ΦLBO
1
0 10 20 30 40 50 60
30 Experimental
Number of events/s
Simulated
10
5
2
1
0 10 20 30 40 50 60
Time (s)
Figure 75. Experimental and simulated number of events in a one second interval
for a transient in equivalence ratio in the LDI combustor.
118
6.4. Time Response Improvement in Engine Decelerations with LBO Margin
Sensing
As described previously, LBO can occur during deceleration transients in gas
turbines due to excessively lean equivalence ratios resulting from the slow spool down
response of engines. To prevent LBO during deceleration transients, current aircraft
engine control systems commonly use a passive method that is based on placing a
minimum limit on the Ratio Unit (RU) [75, 76]. As outlined in Section 2.5, RU is defined
as Wf/Ps3, where Wf is the fuel flow rate and Ps3 is the combustor inlet static pressure.
Thus RU is an approximate measure of a combustor’s overall equivalence ratio (see
Appendix A). Current engine control systems likely use an excessively high margin for
the RU limiter, because it only provides an approximate measure of the equivalence ratio,
and because blowout limits are inherently uncertain. Such excessive margins limit the
transient response of an engine during a deceleration.
If LBO margin sensors are employed, it can be speculated that the RU limiter
might not be needed; instead, LBO prevention would be based on output from the
margin sensors. However, such an approach could result in pushing the combustor
through flame extinction and re-ignition events during each deceleration, which might not
be preferable from an engine performance point of view. Therefore, the nominal LBO
prevention approach based on the RU limiter could still be used, however with a lower
LBO margin. Protection from LBO would be achieved with an LBO margin sensor-
based controller to deal with infrequent near-LBO excursions. The potential advantage of
this approach is that with a lower LBO margin used for the RU limiter, the transient
response during decelerations can be improved.
In order to investigate the potential for improvements in transient response, along
with other issues in LBO margin sensing, a turbofan engine simulation was implemented
using the Numerical Propulsion System Simulation (NPSS) tool. The engine model and
control scheme are described in Section 3.3 of the thesis. For the nominal engine control,
119
without LBO margin sensing, the RU limiter is set at 15.12 lb/hr/psi, which is slightly
below the idle RU of 17.6 lb/hr/psi. To relate this to a standard combustion parameter, the
combustor primary zone equivalence ratio corresponding to the nominal RU limit is
about 0.58, assuming 27% of the total combustor air passes through the primary zone.
The LBO precursors are modeled using the statistical methods described in the previous
sections, and the statistics are based on the LDI combustor precursor results.
6.4.1. Transient Response Improvement
For the nominal case (without margin sensing), the LBO margin limit is assumed
to be Φ/ΦLBO=1.35. This is a reasonable assumption based on the observation that in the
LDI combustor events start to occur at Φ/ΦLBO=1.20, and one would expect the nominal
margin to be where local extinctions are unlikely to occur. Based on this, the LBO
equivalence ratio would be 0.43. However, the LBO limit cannot be a perfectly known
value; it should be uncertain due to variations in temperature, strain, mixing, etc. So if
the nominal RU limit is intended to reliably prevent blowout excursions, one can assume
that the probability that LBO might occur at this nominal equivalence ratio is quite small.
For example, one might expect LBO might occur at the margin limit only once in a
million decelerations. Furthermore for this exercise, we assume that the equivalence ratio
where LBO occurs follows a Gaussian distribution due to the random fluctuations in
operating conditions. Using a probability of 10-6 for LBO occurring at the nominal limit
equivalence ratio and the expected (average) LBO equivalence ratio of 0.43, the
probability of LBO occurring at a particular equivalence ratio can be calculated. The
Gaussian distribution of the LBO equivalence ratio and its CDF are shown in Figure 76.
The next step in the simulation is to choose a new, lower margin (minimum
allowed equivalence ratio) for the case with LBO margin sensing. If successful, the
control system should prevent the occurrence of LBO with the same reliability as the
nominal margin without control, i.e., a probability of blowout occurring that is less than
120
10-6. For this exercise, we choose the new limit as the point where the likelihood of LBO
occurring is ~5%. The resulting minimum allowed equivalence ratio is 0.49. In the figure,
the minimum allowed equivalence ratios for the nominal case (black circle) and the
modified case (red circle) are shown. The reduction in minimum allowed equivalence
ratio results in a corresponding reduction in the RU limiter from the nominal
15.12 lb/hr/psi to 13.32 b/hr/psi (a 12% reduction). Whether such a reduction results in
appreciable improvement in transient response needs to be investigated.

0
15 10
-2
10
Probability density
10
Probability
-4
10
5
-6
10
-8
0 10
0.3 0.43 0.49 0.58 0.3 0.35 0.43 0.49 0.58
LBO Equivalence ratio (Φ) LBO Equivalence ratio (Φ)
(a) (b)
Figure 76. (a) Gaussian PDF of LBO equivalence ratio; (b) probability of LBO
equivalence ratio being above a given equivalence ratio.
For the modified engine case, the RU limit is not suddenly reduced but is
gradually reduced over a 2 s period after the nominal (original) RU limit is reached. A
gradual reduction is necessary since the equivalence ratio has to be reduced somewhat
slowly, in order to allow time for precursor events to occur. The RU traces for nominal
and modified control, for a full power to idle transient, are shown in Figure 77. In Figure
78 corresponding thrust response at sea level static conditions is shown. The time
required to reach 90% idle thrust level for the nominal limiter case is 6.6 s (measured
from the initiation of the transient). For the modified limiter, the required time is reduced
to 4.4 s, providing a 33% improvement transient engine response.
121
`
32
Ratio Unit (RU) (lb/hr/psi)

Nominal
28
Modified
24
20
16
12
0 2 4 6 8 10 12
Time (s)
Figure 77. Engine Ratio Unit (RU) during a full power to idle transient for nominal
and modified RU limit control cases.
35000
Nominal
25000 Modified
Thrust (lbf)
15000
8000
4916
4000
0 2 4 6 8 10 12 14
Time (s)
Figure 78. Thrust response for full power to idle transient at sea level static
conditions with the nominal RU limiter and the modified RU limiter.
As noted before, the modified limiter approach should provide the same low
probability of LBO occurring (<10-6) as the nominal limiter case. For the modified case,
the LBO margin sensor control is assumed to work if it can detect a precursor before the
LBO limit is reached. Therefore, the probability of failure is based on the lack of an LBO
precursor occurring before the transient reduction in equivalence ratio reaches the LBO
122
limit, which as noted above is also assumed to have a Guassian probability distribution.
The LBO equivalence ratio probability and the probability of event occurrence can be
combined in order to estimate the failure probability of the precursor event-based control
system. For a transient in Φ, the probability of failure is the integrated product of LBO
being at a given Φ and the probability that no event occurs before that Φ is reached, as
given by Eq. (6.6).
2
Pfailure   P     P
1
LBO noevents   d  (6.6)
This failure probability is calculated for the modified limiter case above, based on
the simulated equivalence ratio transient. The probability of no events is calculated using
a similar approach as employed in Section 6.3, assuming LBO exists at a given Φ. For the
calculation, different event rate trends obtained using different threshold settings for the
LDI combustor data are used. The resulting failure probabilities and event rates are
shown in Figure 79. Cases 1, 2 and 3 correspond to event rates obtained with lower
thresholds of μ-3.5σ, μ-3.25σ and μ-3.0σ respectively. Recall that the less stringent
threshold produces higher detected event rates. In the figure, the probability of failure is
normalized with the required nominal probability of 10-6. The failure probabilities are
quite sensitive to the event rates. While Case 2 nearly meets the required “safety”
constraint (10-6), Case 1 results in a system that is 100 times more likely to blowout
compared to the nominal case. Similarly, Case 3 exceeds the requirement by more than 3
orders of magnitude. However, it should be noted that the events detected based on the
threshold of Case 3 would likely include some false events.
123
The failure probability calculation approach could be used as guidance for
selection of RU limiter when an active LBO control, based on precursor events, is
employed. The approach used in this example calculation provides a scheme for
estimating the reliability of the margin sensing in order to prevent LBO.
Case1 : -3.5
Normalized probability
30 1.E+02
Case2: -3.25
Number of events/s
Case3: -3.0 1.E+01

20
1.E+00 Case3
Case1 Case2
1.E-01
10
1.E-02
1.E-03
0
1 1.2 1.4 1.6 1.E-04
Φ/ ΦLBO
(b)
(a)
Figure 79. (a) Average event rates for different lower threshold settings; (b)
probability of failure to sense LBO for different event rate trends.
6.4.2. Event based active control
The above analysis addresses whether precursors occur before LBO in a
deceleration transient. It does not address the possible reduction in transient response due
to precursor event occurrence and the control action in response to the events. In order to
investigate this possible reduction in transient response, an additional controller
responding to precursor events is added to the simulation. The control also simulates the
practical scenario, where LBO is controlled based on precursor detection.
For the simulation, LBO is assumed to be at Φ=0.51, above the minimum
equivalence ratio attained during the modified RU limiter case (Φ=0.49). Events are
simulated in real time, similar to the simulation shown in Figure 75, using the μ-3.25σ
threshold event rates. Possible control actions in response to event occurrences are: 1)
increase the fuel flow for a brief period in order to increase equivalence ratio, and 2)
124
maintain a constant fuel flow for a brief period, to allow the air flow rate to drop, thereby
increasing the equivalence ratio. For this simulation, the constant fuel flow option is used.
The event based controller commands the fuel flow rate to be paused for about 100 ms
when an event is detected. A maximum comparator approach is used, i.e., the maximum
of the fuel flows calculated by the event based controller and the regular engine controller
is provided as the command to the engine.
Equivalence ratio responses with the event based controller and without the event
based controller (modified RU limiter case) are shown in Figure 80. With the controller
responding to precursor events, the combustor is prevented from going below the LBO
limit. Events simulated in real time and the equivalence ratio response (normalized as
Φ/ΦLBO) are shown in Figure 81 on an expanded scale. The pause in the fuel flow
reduction clearly prevents the combustor from going below the LBO limit. The number
of events in the simulation is about 13 over the three seconds that the combustor operates
near the LBO limit.
The thrust response for the modified RU limiter control with the added event
based controller is shown in Figure 82, along with a comparison to the case without the
event based controller (presented earlier). The thrust response with the event based
controller is nearly the same as the case with only the modified RU limit, just marginally
slower. The response is not reduced even in the scenario where the LBO limit is above
the assumed RU margin limit. For cases where LBO is below the minimum equivalence
ratio, the thrust response would be nearly the same. Thus the addition of an event based
controller should not significantly affect the transient response improvements estimated
using the simulation with just the modified RU limit.
125
1.8
1.6
1.4 W ithout event controller
Equivalence ratio (Φ)

1.2 W ith event controller
1
0.8
0.6
0.51 LBO limit
0.4
0 2 4 6 8 10
Time (s)
Figure 80. Equivalence ratio response for the modified RU limiter case and event
based control case assuming ΦLBO=0.51.
1
Events
Events
0
0 2 4 6 8 10
1.4 / 
LBO
Φ/ΦLBO
1.2
1
0 2 4 6 8 10
Time (s)
Figure 81. Simulated events in real-time (top) and normalized equivalence ratio
(bottom) for event based control.
126
35000
W ithout event controller
25000 W ith event controller
Thrust (lbf)
15000
8000
4916
4000
0 2 4 6 8 10 12 14
Time (s)
Figure 82. Thrust responses for the modified RU limiter case and with event based
control case.
6.5. Summary
Precursors are discrete events and were observed to occur randomly with wide
variations in the time between successive events. The random nature indicates that for a
sufficiently fast transient that moves the system towards LBO, precursors may not occur
before the flame is lost. Therefore an analysis is required for estimating the probability
that at least one precursor occurs before LBO during a power transient.
For both the LDI and gas-fueled combustor data, the time between successive
precursor events (TBE) is well modeled with an exponential distribution, except for a
slight mismatch at smaller time scales, possibly due to the finite duration of precursors,
which is not included in the theoretical distribution. The exponential distribution for
TBE implies that precursor event occurrences can be modeled with Poisson statistics.
Poisson statistics can thus be used to calculate the probability of at least one precursor
occurring before LBO during a transient. This method provides a basis for determining
the maximum allowed fuel flow reduction rate that can be implemented without risking
flame loss. Moreover, it was demonstrated that discrete event simulation approaches can
127
be used for simulating precursor events in real-time. This enables simulation and analysis
of LBO control approaches based on precursor event detection.
Results presented here show that LBO margins can be reduced during
deceleration transients in turbine engines, thus improving transient response. The possible
improvements in transient response are examined in a model turbofan engine simulation.
The results indicated considerable improvement in thrust response (33%) by a reasonable
reduction in the LBO margin (12%). An approach has been proposed for providing a
basis for LBO margin reduction while implementing margin sensors, such that the margin
sensing is reliable to the same degree as provided by the higher LBO margin without
sensing and control. An active LBO control approach simulating precursor event based
LBO control in engines during deceleration transients was also demonstrated.
128
CHAPTER 7
CONCLUSIONS AND RECOMMENDATIONS
This chapter summarizes the findings of this thesis, describes the potential impact
of the results, and provides recommendations for future work in the area of LBO
precursor sensing and control.
7.1 Summary and Conclusions
The four primary objectives of this thesis and the approach for each are listed
below. Following that, more detailed conclusions are presented.
Robust LBO Margin Sensing
1) To investigate LBO margin sensing under dynamically unstable conditions,
as this issue has not been addressed in previous studies. This was accomplished
by investigating LBO margin sensing in a premixed gas-fueled combustor that
exhibited pronounced combustion dynamics.
2) To investigate LBO margin sensing under combustor operating conditions

that are more representative of practical combustors, i.e., at elevated pressure
and preheat temperature operation. This has been addressed by investigating
margin sensing in a liquid-fueled, LDI-type combustor designed for low NOx
emissions and operating at elevated pressure and temperature.
LBO Margin Sensing for Rapid Transients
3) To analyze the limitations of LBO margin sensing approach in rapid

transients leading to LBO. This has been accomplished by developing a
129
probability model and analysis approach for the time between the random
precursor events.
4) To address the benefits and issues in implementing the LBO margin sensing
in deceleration transients of aircraft engines. This issue has been addressed by
simulating the transient response of a turbofan engine with the addition of
probability models for the LBO limit and the appearance of precursors during the
transient.
7.1.1. Robust LBO Margin Sensing
In previous studies, LBO margin sensing was mainly studied under dynamically
stable conditions. As dynamic instability can considerably alter flame and flow fields, it
was initially uncertain if the previously observed LBO precursors would exist or be
detectable in optical and acoustic signals dominated by high amplitude instability. To
address these issues, margin sensing was studied in a gas-fueled combustor configuration
with a centerbody, representative of ground-based gas turbine combustors. Through
appropriate choice of combustor length, the system could be made to exhibit strong
dynamic instability. In order to examine margin sensing under different instability
scenarios, the combustor was capable of operating in a premixed mode with fuel added
far upstream, and a mode with fuel addition just upstream of the dump plane, providing
dynamics that included equivalence ratio oscillations mechanism.
The results of Chapter 4 demonstrated that precursor events still exist in the
presence of dynamics, and their characteristics are not significantly altered by the
presence of dynamics. Though precursors could be detected in the raw optical signals
with large scale dynamic fluctuations, different event detection approaches were required
for the different instability modes. Low-pass filtering of the signal, however, was shown
to suppress the dynamics in the signal and provide the same precursor signature for all
the three cases: dynamically stable, and unstable with and without equivalence ratio
130
oscillations. The universal signature allowed a single precursor detection algorithm to
work regardless of the dynamic stability characteristics of the combustor. This is
important because the dynamic characteristics in fielded engines can change with
operating conditions. Therefore, having one common detection algorithm greatly reduces
the complexity of the detection software.
Precursor events detected in the optical signals were observed to increase in
number as LBO conditions were approached, providing a clear measure of the proximity
to LBO. On average, the event occurrence rate near LBO was observed to be in the range
of 1-2 per second and the event durations were in the range 20-50 ms. A novel parameter
defined in this thesis, the Stability Index (SI), combines event rate, event duration and
event strength to produce a more robust LBO proximity parameter that provides a higher
dynamic range between statically stable and nearly unstable operating conditions.
The same LBO precursor detection approach was applied to a different combustor
configuration, a liquid-fueled, LDI combustor operating at elevated pressure and preheat
temperature. Near LBO, the combustor exhibited partial extinction and re-ignition events.
These events were observed to have much shorter durations (1.5-3 ms) compared to the
gas-fueled combustor. In addition, this combustor produced a higher average event rate
near LBO. The experimental data from the gas-fueled combustor show that the LBO
precursors correspond to the flame temporarily switching from a compact flame mode
with burning around the inner recirculation zone to a long flame with most of the burning
taking place downstream. In the LDI combustor, on the other hand, no such flame mode
change was observed; rather the precursors appear to be local extinctions in the flame
close to the injector.
The LDI combustor operated with a moderate level of dynamic instability
occurring at multiple frequencies. The period of the dynamic instability (1.4-1.7 ms) was
similar to the duration of the precursor events in the LDI combustor. While events were
sometimes observable in the raw optical signals, comparison to high speed imaging
131
suggested other events could be obscured by dynamic oscillations and high frequency
noise. Once again, low-pass filtering enabled robust event detection. In the optical
signals, CH* chemiluminescence showed improved performance compared to OH*
chemiluminescence, producing more events near LBO. Results from this combustor were
also used to demonstrate the capability of detecting LBO margin in real-time.
LBO margin sensing using acoustic signals was also investigated in both
combustors. Under dynamically unstable conditions in the gas-fueled combustor, no
apparent precursor signature was observed in the raw acoustic signals. However, by low-
pass filtering the signals precursor signatures became evident. On the other hand in the
LDI combustor, neither the raw signals nor the low-pass filtered signals revealed
precursor events. One likely reason is the smaller heat release changes during precursors
due to the lack of a mode change. In addition, it was demonstrated that even if precursor
event durations and dynamic instability period have nearly the same time scales, low-pass
filtering would be less effective for separating precursors from dynamics with acoustic
detection. Finally, the effect of pressure reflections from combustor boundaries was also
addressed by modeling reflections in a one-dimensional combustor geometry with an
impedance jump caused by the flame. The results indicated that reflections cause ringing
in the detected acoustic precursor signal. The ringing is at the combustor’s natural axial
mode frequency, which is often the same as the frequency of the dynamic instability. This
manifestation makes it hard to separate precursors from combustion dynamics. This
ringing is not a significant issue if the event duration is long compared to the round-trip
travel time, which was the case for the gas-fueled combustor. It is recommended that for
improved detection of precursors in the presence of dynamic instability, pressure
transducer should be located at combustor exit where dynamic instability amplitudes are
low, thus minimizing the affect on precursor signal.
132
7.1.2. LBO Margin Sensing in Rapid Transients
Precursor events are discrete and occur randomly in time with wide variations in
time between events. Therefore for a sufficiently fast transient leading to LBO,
precursors may not occur before blowout occurs. Hence, transients would have to be
limited to a rate that allows at least one precursor to occur before LBO.
It has been shown that time between successive events can be modeled reasonably
well with an exponential distribution. This indicates the event occurrence can be
described by a Poisson process, or the probability of a precursor event occurring during
some time can be calculated from a Poisson distribution. This distribution was used to
calculate the probabilities for at least one event to occur for a transient in equivalence
ratio, thus providing a basis for determining the limiting transient fuel reduction rate. It
has been demonstrated that discrete event simulation approaches could be used for
simulating precursor events in real-time. An illustrative test simulation matched closely
with the experimental results. Such simulation capability will aid in LBO control
simulations using precursor events, without having to do costly experiments.
The potential for transient response improvement, during decelerations using
LBO margin sensors has been demonstrated with a simulation of a turbofan engine. The
results indicated considerable improvement (33% for the example presented) in thrust
response by a reasonable lowering of LBO margins (12%) while employing LBO margin
sensors. An approach has been proposed for providing a basis for LBO margin reduction
while implementing margin sensors, such that the margin sensing is reliable to the same
degree as provided by the higher LBO margin without sensing and control. An event-
based, active LBO control approach for combining with existing RU limit controllers was
also described and implemented in the simulation.
133
7.2. Recommendations for Future Work
The precursor events in the gas-fueled combustor are associated with a flame
mode shift, whereas in the LDI combustor no such shift was observed. This is an
indication that not all combustors will have flame mode changes and hence precursor
characteristics could vary greatly between combustors. A better knowledge of the flame
and flow field dynamics near LBO are required in order to explain the observed
differences and thus provide the ability to predict the expected precursor characteristics
for different combustors.
An instantaneous flame image obtained during a representative long flame mode
(precursor) in the gas fueled combustor is shown in Figure 83. This flame mode consists
of a large downstream flame, burning most of the fuel and flame along two helical like
regions. Besides the flame zones shown in the figure, a small flame near the inlet above
the center body is often observed. The helical regions are most likely precessing vortex
cores (PVCs) as they were observed to rotate about the combustor geometrical axis. In
addition, they exist for a distance of about 1.5 times the combustor diameter, from the
inlet, similar to PVCs. These helical flames could be igniting the un-burnt fuel
downstream and stabilizing the downstream flame, creating a long flame mode. The
PVCs could favor the presence of a flame due to relatively smaller strains rates
experienced there, as was observed in a study by Stohr et al. [54]. Hence, it is possible
that existence of PVCs is primarily responsible for causing a flame mode change in the
gas fueled combustor. In the LDI combustor PVCs might not be present, due to its
geometric features, confinement, and combustion mode, thus not resulting in a flame
mode change.
To ascertain the existence of PVCs in the gas fueled combustor during precursors
high speed PIV in a horizontal plane could be performed. Addition of simultaneous PLIF
134
(OH or CH) to PIV would further corroborate the existence of flame along PVCs. Similar
experiments in the LDI combustor would show the absence of PVCs.
Downstream
flame
Helical flame
Figure 83. An example flame configuration during precursor events from a

time resolved chemiluminescence image.
Combustion dynamics, mainly at low frequencies, growing in amplitude and
causing blow-off is often observed in gas turbine engines. Though the current study
investigated LBO margin sensing in the presence of dynamics, it does not precisely
address the scenario of transient growth in dynamics causing blow-off. In the current
study the dynamics were at their limit cycle amplitudes and did not exclusively cause
blow-off. The possibility of precursors occurring before blow-off could be limited by the
rate at which instabilities grow before resulting in blow-off, similar to the equivalence
ratio reduction transients investigated in this study. LBO margin sensing for the scenario
of transient growth in dynamics resulting in blow-off is worth further investigation. The
transient growth could be simulated by employing a fuel system tuner (FST) to initially
damp the dynamics and later letting the dynamics amplitude grow by changing the FST
properties.
The acoustic precursor reflection model used in the current study could be
improved further by including transmission losses at wall boundary layers. Such analysis
would give a more accurate waveform for reflected precursor signature. Instead of the
approach used in the current study for simulating reflections, alternative approaches for
simulating impulse response of the combustor could be used. For example, a layer
135
peeling algorithm, which is often used for simulating impulse response of a wave tube
with multiple impedance jumps [84, 85], might be appropriate. With this approach, a
larger number of impedance discontinuities, reflections, and, transmission losses could be
implemented. In addition, the possibility of deconvolving the impulse response from the
measured acoustic signal could be investigated in order to reveal the actual precursor
signature.
In the current study, the exponential distribution hypothesis for time between
successive events is validated using Kolmogorov-Smirnov test. Alternative hypothesis
testing methods such as Shapiro-Wilk test or Anderson-Darling test could be employed
for more rigorous validation. The current study demonstrated a method for calculating the
probability for a given number of precursors to occur during a transient before LBO is
reached. The analysis assumes that the combustor precursor statistics during a transient
are the same at each condition (equivalence ratio, mass flow rate, etc.) as for the
corresponding steady-state operating point. The validity of this assumption, which may
fail, for example, due to thermal inertia effects, should be explored. In addition, the
theoretical results could be validated by conducting several fast transients in a combustor.
The discrete event simulation algorithm used in the current study for simulating precursor
events in real-time can be refined by including a constraint on the minimum allowed time
between two successive events in order to account for the finite duration of precursor
events. Besides, alternative simulation algorithms could be explored for more accurate
simulation of precursor events.
The active LBO control approach demonstrated in the current study could be
refined further with more realistic implementation. For example precursor event
occurrence rates could be scheduled with combustor operating conditions since the
precursor properties can be expected to be dependent on the operating conditions. In
addition instead of a constant RU limiter a varying RU limiter could be used, as LBO
fuel-air ratio can change with the operating conditions. Besides, alternative control
136
methods to prevent LBO such as increasing fuel flow rate when an event is detected or
selecting a minimum RU limit when the first event is detected could be explored. The
deceleration transient response improvements observed in the current study could be
validated by implementing margin sensing and active LBO control in a real engine. In
addition to aircraft gas turbines, transient response improvements could be investigated in
land based gas turbines during rapid load shedding incidents.
Though LBO margin sensing was successful in the combustors employed in the
current study, the same many not be true for other combustors. The ability to detect
precursor events depends on the strength of precursor events compared to the noise in the
optical signals. Combustors with significant combustion unsteadiness could produce
sufficiently high noise levels to prevent precursor detection. In addition, precursors
should have sufficient signatures, i.e., large enough flame extinctions, in order to detect
them. The spatial extent of extinction before re-ignition could be dependent on several
features such as mean flow field and turbulence characteristics, non-premixedness,
stratification in the mixture, and fuel spray characteristics. Such dependency can result in
wide variations in the extinction and re-ignition processes between combustors. A more
detailed study of dependency of extinction and re-ignition event characteristics on flow
field and fuel distribution is required in order to generalize our understaning of the
characteristics of precursors.
137
APPENDIX A
ENGINE FUEL-AIR RATIO ESTIMATION
In gas turbine engines fuel flow rate is measured directly whereas the combustor
air flow rate cannot be measured. However, an estimation of combustor fuel-air ratio is
required for preventing lean blowout and avoiding turbine over temperature etc. Flow
through the turbine inlet nozzle (i.e, combustor exit) is choked over most of the operating
range of an engine [74]. Therefore the mass flow rate through the nozzle, assuming
choking is given by Eq. (A.1), where A* is the nozzle area, K is a constant and the
subscript 4 denotes combustor exist conditions. Though the equation gives a means for
calculating the flow rate, combustor exit flow properties (P04 and T04) are not usually
measured, due to the harsh environment and instead combustor inlet properties are
measured.
 1 2( 1)
P04  2  P04
m4  A *
  K
  1 
(A.1)
RT04 T04
The stagnation flow properties at the nozzle are nearly same as the stagnation properties
sufficiently upstream of the nozzle, inside the combustor. Due to small mach numbers
inside the combustor the stagnation properties can be approximately replaced by static
properties, P4 and T4. Further, due to a small pressure drop across the combustor inlet,
combustor pressure can be approximated to compressor exit static pressure, i.e, P4  P3 .
These approximations give rise to Eq.(A.2) for the mass flow rate.
P3
m4  K (A.2)
T4
138
In combustors usually the amount of fuel is a small fraction of the airflow rate, and thus
m4  ma . The ratio of fuel flow rate to air flow rate is given by Eq. (A.3).
mf mf
K T4 (A.3)
ma P3
Since the temperature of combustion products (T4) is essentially is a function of fuel-air
air ratio, we obtain
mf m 
 fn  f  (A.4)
P3  ma 
Thus the ratio of fuel flow rate to the compressor exit static pressure (Ratio Unit,
i.e, RU) provides a measure of combustor fuel-air ratio. However, the equation (A.4) does
not indicate a linear relationship between RU and fuel-air ratio, due to the presence of
T4 term. However over the operating range of an engine, the relative change in T4 is
quite small where as P3 changes by an order of magnitude [74]. This implies that air flow
rate is approximately proportional to P3 in Eq. (A.2) and hence RU is approximately
linearly proportional to fuel-air ratio.
139
APPENDIX B
LDI FUEL NOZZLE TESTING
The fuel supply tube section of the fuel nozzle unit supplied by NASA was
modified in order to fit the setup used in the study. The original fuel supply tube had an
L-shaped bend with a very short straight section between the bending location and the
nozzle stem (see Figure 84). However the current LDI injector required a much longer
straight section before the bend to fit into the facility. In addition, it was required to
increase the tube diameter slightly (from 2.28 mm OD to 3.17 mm) in order to use
standard compression tube fittings. Therefore, the fuel supply tube and the tube providing
the air gap were cut, leaving a short straight section, on to which straight tubes with
slightly higher diameter were welded (see Figure 84 ). At the joining location there is a
sudden change in area that may cause cavitation. Because of the modifications, it was
required to test the original and modified nozzles for similar spray performance. The
spray produced by each was characterized with qualitative spray imaging and Phase
Doppler Particle Anemometry (PDPA).
Figure 84. The original bent fuel nozzle and the modified straight nozzle.
Spray visualization was performed in order to obtain qualitative spray structure,
cone angle, etc. These experiments were carried out with pressurized water and with the
140
spray injected into quiescent room air. The visualization was performed by illuminating
the spray with a laser sheet and taking images perpendicular to the sheet. The laser, a
copper vapor laser (Meta Laser Technologies Inc, MLT-20), was operated with a pulse
rate of 5.7 KHz with a pulse duration of 5 ns. The nozzles were tested with different
pressure differentials in the range 30-75 psi. Example spray images obtained at 70 psi for
both the nozzles are shown in Figure 85. The images show similar spray structure for
both the nozzles. In addition, the spray cone angles are around 40° for both. The spray
forms into a conical sheet near the nozzle orifice and subsequently breaks down into
droplets. Spray images at other pressure differentials indicated similar characteristics for
both nozzles.
0 0
2 2
4 4
39°
40°
6 6
8 8
10 10
cm cm
Figure 85. Spray images for (left) bent nozzle (right) straight nozzle at 70 psi pressure
differential.
Spray characterization with a 2-D PDPA system was performed to measure
droplet sizes and velocities. For these experiments, the pressure differential was
maintained at 70 psi. Measurements were carried out in a plane perpendicular to the spray
axis along a radial line. Drop diameters expressed as Sauter Mean Diameters (SMD or
141
D32) for axial locations 20 and 50 mm from the nozzle tip are plotted in Figure 86, for the
bent and straight nozzles. A radial position of zero corresponds to the spray center line.
The figure conveys similar drop sizes for both nozzles. From the figure, SMD are in the
range 30-80 µm, indicating that the nozzles are designed to produce a moderately fine
spray.
Droplet average velocities in the axial direction are shown in Figure 87 for two
axial locations. Both nozzles have similar axial velocities, and the velocity profile in the
radial direction changes with axial location. Near the nozzle tip, there is a high velocity
along the cone edges and lower velocities along the centerline. Farther downstream, the
velocity peaks along the center line and decreases with radial distance.
80 90
Sauter mean diameter (µm)
Sauter mean diameter (µm)

70 80
60
70
50
40 60
Bent
30 50 Bent
Straight Straight
20 40
-2 -1 0 1 2 -3 -1 1 3
Radial position (cm) Radial position (cm)
Figure 86. Droplet sauter mean drop diameters (left) at an axial location of 2cm
(right) at an axial location of 5cm, from nozzle tip, for bent and straight nozzles.
142
6 5
Average velocity (m/s)

5 4
Average velocity (m/s) 4
3
3
2
2 Bent
Bent
1 Straight 1
Straight
0 0
-2 -1 0 1 2 -3 -1 1 3
Radial position (cm) Radial position (cm)
Figure 87. Droplet average axial velocities (left) at an axial location of 2cm
(right) at an axial location of 5cm, from nozzle tip, for bent and straight nozzles.
143
APPENDIX C
LDI COMBUSTOR DEVELOPMENT
The present study used a single element LDI injector derived from a 9-element
LDI injector developed by NASA Glenn Research Center [5]. The single element version
uses a co-flow around the LDI element flow in order to compensate for the absence of
surrounding elements. Since this was the first time the single element configuration has
been operated with a co-flow, great attention was devoted to ensuring its operation was
similar to the expected behavior of the multi-element LDI system. Specifically, the single
element LDI injector should have a lean partially premixed flame, with low emissions,
similar to the full nine-element NASA LDI injector. In addition, similarity in qualitative
flame shape and flame size to the multi-element injector is desired, as practical
combustors would employ multi-element configurations.
To provide confinement for the central LDI element, similar to the confinement
by surrounding elements in a multi-element configuration, a co-flow is used. The co-flow
is created using a perforated plate. The mass flow rate split between the co-flow and the
LDI element flow determines the effective confinement. In the original nine element
injector, the flow split ratio is eight (between surrounding elements and the central
element), inside a square cross section of 76.2 mm on each side. For the single element
injector, sitting in a test section with a circular cross section and a 76.2 mm nominal
diameter, the flow split ratio that would produce the same average axial velocity ratio (in
the absence of combustion) between the center element and the surrounding flow drops to
6.3. However, this does not account for any loss of confinement associated with the
absence of swirl in the surrounding flow.
144
A perforated plate with 2.38 mm diameter holes, uniformly distributed across the
plate, with an overall area blockage of 88% was designed in order to produce the required
flow split ratio of 6.3. However, effective area measurements of the injector and
perforated plate indicate the actual flow split ratio was closer to 8. Combustion testing of
the injector with this plate produced a “clean” blue flame at atmospheric pressure, but an
orange flame at elevated pressures (> 2 atm), as shown in Figure 88. In the figure, the
velocity is the average un-burnt flow velocity and the temperature is the inlet air
temperature. Equivalence ratio is the overall equivalence ratio, calculated from the total
fuel and air flow rates. The existence of a blue flame is an indication of lean
premixed/partially premixed operation, whereas the existence of an orange flame
indicates formation of soot, typical for non-premixed operation. In addition, the blue
flame at ~1 atm can be observed to spread to almost the entire width of the combustor,
indicating that the co-flow produced less flame confinement than expected from a multi-
element injector, where the flame spread would typically be limited to the size of a single
element (25.4 mm). Moreover, the injector produced a twin flame structure, seen in the
blue flame at atmospheric pressure, having two distinctively bright flames. Thus
improvements to the injector design were required in order to reduce the flame spread
and to produce a more premixed (blue) flame at elevated pressures.
In addition to the flow split ratio affecting the flame spread, fuel nozzle tip
position relative to the throat of the venturi has also been observed to affect the flame
spread significantly. For the images presented in Figure 88, the fuel nozzle tip was
located slightly ahead of the venturi throat, by a small distance of ~1 mm. However
placing the nozzle tip slightly behind the venturi throat, at a distance of 1.5 mm, produced
a significant decrease in flame spread at atmospheric pressure, as shown in Figure 89. In
addition, the twin flame structure observed earlier disappeared. Even with this
improvement, there was a slight appearance of orange trails in the atmospheric tests,
145
indicating excessive non-premixedness, and an orange flame was observed at elevated
pressures.
Figure 88. Flame in the LDI combustor for (a) P=1.17 atm, T=663 K, V=14 m/s,
Φoverall=0.5 (b) P=2atm, T=682K, V=10.6m/s, Φoverall =0.3.
Figure 89. Flame in the combustor at P=1.13 atm, T=654K, V= 13.4m/s, Φoverall=0.32,
for fuel nozzle position upstream of venturi throat.
Improved operation was obtained by reducing the flow split ratio close to 6.3. For
a given average velocity in the combustor, the reduced flow split ratio results in a higher
velocity through the LDI element. Higher air velocity through LDI element’s venturi
146
throat would improve spray break up creating finer droplets promoting premixed
combustion. However, a reduction in flow split would decrease confinement by the co-
flow, as its mass flow rate decreases. To compensate for this, the co-flow velocity profile
was modified such that there is a higher velocity surrounding the center and lower farther
away. To achieve such a profile, some of the holes away from the center were blocked.
Besides the flow split modification, the fuel nozzle tip was positioned behind the venturi
at a distance of ~1 mm. This arrangement produced a blue flame at atmospheric pressure
and at elevated pressures, as shown in Figure 90. In addition, the flame spread is
reduced, suggesting improved confinement of the fuel/flame by the higher velocity co-
flow. All the LBO sensing results in this thesis correspond to combustor operating in this
final configuration.
Figure 90. Flame in the combustor at (a) P=1atm, T=663K , V=14 m/s, Φoverall=0.41
(b) P=2atm, T=682K, V=10.6m/s, Φoverall=0.28 (c) P=4atm, T=733K, V=12m/s,
Φoverall=0.23 .
147
REFERENCES
1. Bowman, C. T. "Control of combustion-generated nitrogen oxide emissions:

technology driven by regulation," Symposium (International) on Combustion Vol. 24,
No. 1, 1992, pp. 859-878.
2. Penner, J. E., Lister, D., Griggs, D. J. Dokken, D. J. and McFarland, M.,

"Aviation and the Global Atmosphere: A Special Report of the Intergovernmental Panel
on Climate Change." Cambridge, UK., 1999.
3. Sanjay, M. C. "A review of NOx formation under gas-turbine combustion

conditions," Combustion science and technology Vol. 87, No. 1-6, 1993, pp. 329-362.
4. Gokulakrishnan, P., Ramotowski, M., Gaines, G., Fuller, C., Joklik, R., Eskin, L.,
Klassen, M., and Roby, R. "A Novel Low NO Lean, Premixed, and Prevaporized
Combustion System for Liquid Fuels," Journal of Engineering for Gas Turbines and
Power Vol. 130, No. 5, 2008, p. 051501.
5. Tacina, K. M., Lee, C. M., and Wey, C. " NASA Glenn High Pressure Low NOx
Emissions Research." NASA/TM-2008-214974, 2008.
6. Feitelberg, A. S., and Lacey, M. A. "The GE rich-quench-lean gas turbine

combustor," Journal of engineering for gas turbines and power Vol. 120, No. 3, 1998,
pp. 508-508.
7. Lefebvre, A. H. Gas turbine combustion, Taylor & Francis publishers,

Philadelphia, 1998.
8. Lilley, D. G. "Swirl Flows in Combustion - Review," AIAA Journal Vol. 15, No.
8, 1977, pp. 1063-1078.
9. Sung, C., and Law, C. "Extinction mechanisms of near-limit premixed flames and
extended limits of flammability," Symposium (International) on Combustion Vol. 26, No.
1, 1996, pp. 865-873.
10. Mitch Cohen "7F Users Group – preventing blowout trips." Comined Cycle
Journal, 3rd quarter, 2011.
11. Bahr, D. "Technology for the design of high temperature rise combustors,"
Journal of propulsion and power Vol. 3, No. 2, 1987, pp. 179-86.
12. Mongia, H. C. "TAPS–A 4th Generation Propulsion Combustor Technology for

Low Emissions," 2003 AIAA/ICAS International Air and Space Symposium and
Exposition: The Next 100 Years. Dayton, OH, 2003.
148
13. Lenertz, J. E. "Deceleration fuel control system for a turbine engine." U.S.,
Patenet 5596871, 1997.
14. Longwell, J. P., Frost, E. E., and Weiss, M. A. "Flame stability in bluff body
recirculation zones," Industrial & Engineering Chemistry Vol. 45, No. 8, 1953, pp. 1629-
1633.
15. Shanbhogue, S. J., Husain, S., and Lieuwen, T. "Lean blowoff of bluff body
stabilized flames: Scaling and dynamics," Progress in Energy and Combustion Science
Vol. 35, No. 1, 2009, pp. 98-120.
16. Zukoski, E. E. "Afterburners," Aerothermodynamics of aircraft engine

components. AIAA education series, New York, NY, 1985.
17. Hedman, P. O., Fletcher, T. H., Graham, S. G., Timothy, G. W., Flores, D. V.,
and Haslam, J. K. "Observations of Flame Behavior in a Laboratory-Scale Pre-mixed
Natural Gas/Air Gas Turbine Combustor from PLIF measurements of OH," Journal of
Engineering for Gas Turbines and Power Vol. 127, No. 4, 2005, pp. 724-739.
18. Griebel, P., Boschek, E., and Jansohn, P. "Lean Blowout Limits and NO
Emissions of Turbulent, Lean Premixed, Hydrogen-Enriched Methane/Air Flames at
High Pressure," Journal of engineering for gas turbines and power Vol. 129, No. 2,
2007, pp. 404-410.
19. Sturgess, G., Sloan, D., Lesmerises, A., Heneghan, S., and Ballal, D. "Design and
development of a research combustor for lean blow-out studies," Journal of engineering
for gas turbines and power Vol. 114, No. 1, 1992, pp. 13-19.
20. Sturgess, G., Heneghan, S., Vangsness, M., Ballal, D., Lesmerises, A., and
Shouse, D. "Effects of back-pressure in a lean blowout research combustor," Journal of
engineering for gas turbines and power Vol. 115, No. 3, 1993, pp. 486-498.
21. Muruganandam, T., Nair, S., Scarborough, D., Neumeier, Y., Jagoda, J., Lieuwen,
T., Seitzman, J., and Zinn, B. "Active control of lean blowout for turbine engine
combustors," Journal of propulsion and power Vol. 21, No. 5, 2005, pp. 807-814.
22. Nair, S., and Lieuwen, T. "Acoustic detection of blowout in premixed flames,"
Journal of propulsion and power Vol. 21, No. 1, 2005, pp. 32-39.
23. Prakash, S., Nair, S., Muruganandam, T., Neumeier, Y., Lieuwen, T., Seitzman, J.
M., and Zinn, B. T. "Acoustic based rapid blowout mitigation in a swirl stabilized
combustor," ASME Turbo Expo 2005: Power for Land, Sea and Air, Paper GT2005-
68589, Reno-Tahoe, Nevada, USA, 2005.
24. Taware, A., Shah, M., Wu, P., Yang, Y., Zhou, J., and Singh, A. "Acoustics
Based Prognostics for DLE Combustor Lean Blowout Detection," ASME Turbo Expo:
Power for Land, Sea, and Air, Paper GT2006-90231, Barcelona, Spain 2006.
149
25. Muruganandam, T., Nair, S., Olsen, R., Neumeier, Y., Meyers, A., Jagoda, J.,
Lieuwen, T., Seitzman, J., and Zinn, B. "Blowout control in turbine engine combustors,"
42nd Aerospace Sciences Meeting & Exhibit, Paper AIAA-2004-0637, Reno, NV, 2004.
26. Nair, S., Muruganandam, T., Olsen, R., Meyers, A., Seitzman, J., Zinn, B.,
Lieuwen, T., Held, T., and Mongia, H. "Lean Blowout Detection in a Single Nozzle Swirl
Cup Combustor," 42nd Aerospace Sciences Meeting & Exhibit, Paper AIAA-2004-138,
Reno, Nevada, 2004, 2004.
27. Yi, T., and Gutmark, E. J. "Real-time prediction of incipient lean blowout in gas
turbine combustors," AIAA journal Vol. 45, No. 7, 2007, pp. 1734-1739.
28. Thornton, J. D., Straub, D. L., Chorpening, B. T., Huckaby, E. D., Richards, G.
A., and Benson, K. "A combustion control and diagnostics sensor for gas turbines,"
ASME Turbo Expo 2004: Power for Land, Sea, and Air, Paper GT2004-53392, Vol. 2,
Vienna, Austria 2004, pp. 535-543
29. Li, H., Zhou, X., Jeffries, J. B., and Hanson, R. K. "Active control of lean
blowout in a swirl-stabilized combustor using a tunable diode laser," Proceedings of the
Combustion Institute Vol. 31, No. 2, 2007, pp. 3215-3223.
30. Keller, J. J. "Thermoacoustic oscillations in combustion chambers of gas

turbines," AIAA journal Vol. 33, No. 12, 1995, pp. 2280-2287.
31. Lieuwen, T. "Modeling premixed combustion-acoustic wave interactions: A

review," Journal of propulsion and power Vol. 19, No. 5, 2003, pp. 765-781.
32. Snyder, T. S., and Rosfjord, T. J. "Active gas turbine combustion control to
minimize nitrous oxide emissions." United Technologies Corporation, U.S. Patent-
5706643, Jan 13, 1998.
33. Lieuwen, T. "CDMS Helps Prevent Forced Outages, Tune Engine After
Overhaul," Combined Cycle Journal, Third Quarter, 2008.
34. Cohen, J., and Anderson, T. "Experimental Investigation of Near-Blowout

Instabilities in a Lean Premixed Step Combustor," 34th AIAA Aerospace Sciences
Meeting and Exhibit ,Paper AIAA-1996-0819, Reno, NV, 1996.
35. Turns, S. R. An introduction to combustion: concepts and applications: McGraw-

hill publications, New York, 1996.
36. Cheng, R. K. "Low Swirl Combustion," Gas Turbine Handbook, US Dept. of

Energy, 2006.
37. Syred, N., and Beér, J. M. "Combustion in swirling flows: a review," Combustion
and Flame Vol. 23, No. 2, 1974, pp. 143-201.
150
38. Hillemanns, R., Lenze, B., and Leuckel, W. "Flame Stabilization and Turbulent
Exchange in Strongly Swirling Natural Gas Flames," Twenty-First Symposuim
(International on Combustion), Vol. 21, 1988, pp. 1445-1453.
39. Feikema, D., Chen, R. H., and Driscoll, J. F. "Enhancement of flame blowout
limits by the use of swirl," Combustion and Flame Vol. 80, No. 2, 1990, pp. 183-195.
40. Syred, N. "A review of oscillation mechanisms and the role of the precessing
vortex core (PVC) in swirl combustion systems," Progress in Energy and Combustion
Science Vol. 32, No. 2, 2006, pp. 93-161.
41. Leibovich, S. "The structure of vortex breakdown," Annual Review of Fluid

Mechanics Vol. 10, No. 1, 1978, pp. 221-246.
42. Lucca-Negro, O., and O'Doherty, T. "Vortex breakdown: a review," Progress in

Energy and Combustion Science Vol. 27, No. 4, 2001, pp. 431-481.
43. Ruith, M., Chen, P., Meiburg, E., and Maxworthy, T. "Three-dimensional vortex
breakdown in swirling jets and wakes: direct numerical simulation," Journal of Fluid
Mechanics Vol. 486, No. 486, 2003, pp. 331-378.
44. Vandervort, C. "9 ppm NOx/CO combustion system for F class industrial gas
turbines," Journal of engineering for gas turbines and power Vol. 123, No. 2, 2001, pp.
317-321.
45. Escudier, M., and Keller, J. "Recirculation in swirling flow-a manifestation of

vortex breakdown," AIAA Journal Vol. 23, No. 1, 1985, pp. 111-116.
46. Liang, H., and Maxworthy, T. "An experimental investigation of swirling jets,"
Journal of Fluid Mechanics Vol. 525, 2005, pp. 115-159.
47. Wang, S., Hsieh, S. Y., and Yang, V. "Unsteady flow evolution in swirl injector
with radial entry. I. Stationary conditions," Physics of Fluids Vol. 17, No. 4, 2005, p.
045106.
48. Huang, Y., and Yang, V. "Bifurcation of flame structure in a lean-premixed swirl-
stabilized combustor: transition from stable to unstable flame," Combustion and Flame
Vol. 136, No. 3, 2004, pp. 383-389.
49. Bradley, D., Gaskell, P., Gu, X., Lawes, M., and Scott, M. "Premixed turbulent
flame instability and NO formation in a lean-burn swirl burner," Combustion and Flame
Vol. 115, No. 4, 1998, pp. 515-538.
50. Chterev, I., Foti, D., Seitzman, J., Menon, S., Lieuwen, T.,. "Flow Field
Characterization in a Premixed, Swirling Annular Flow," 50th AIAA Aerospace Sciences
Meeting including the New Horizons Forum and Aerospace Exposition, Nashville,
Tennessee, Jan. 9-12, 2012 2012.
151
51. Wicksall, D., Agrawal, A., Schefer, R., and Keller, J. "The Interaction of Flame
and Flow Field in a Lean Premixed Swirl-stabilized Combustor Operated on H2/CH4/air,"
Proceedings of the Combustion Institute Vol. 30, No. 2, 2005, pp. 2875-2883.
52. Ji, J., and Gore, J. P. "Flow structure in lean premixed swirling combustion,"
Proceedings of the Combustion Institute Vol. 29, No. 1, 2002, pp. 861-867.
53. Law, C. K. Combustion physics: Cambridge University Press, Cambridge, UK,

2006.
54. Stöhr, M., Boxx, I., Carter, C., and Meier, W. "Dynamics of lean blowout of a
swirl-stabilized flame in a gas turbine model combustor," Proceedings of the Combustion
Institute Vol. 33, No. 2, 2011, pp. 2953-2960.
55. Turns, S. R. An introduction to combustion: concepts and applications: McGraw-

hill, New York , NY, 1996.
56. Peters, N. Turbulent Combustion: Cambridge University Press, Cambridge, UK,

2000.
57. Ballal, D., and Lefebvre, A. "Weak extinction limits of turbulent heterogeneous
fuel/air mixtures," J. Eng. Power;(United States) Vol. 102, No. 2, 1980, pp. 416-421.
58. Strakey, P., Sidwell, T., and Ontko, J. "Investigation of the effects of hydrogen
addition on lean extinction in a swirl stabilized combustor," Proceedings of the
Combustion Institute Vol. 31, No. 2, 2007, pp. 3173-3180.
59. Hoffmann, S., Habisreuther, P., and Lenze, B. "Development and assessment of
correlations for predicting stability limits of swirling flames," Chemical Engineering and
Processing: Process Intensification Vol. 33, No. 5, 1994, pp. 393-400.
60. Sloan, D. G., and Sturgess, G. J. "Modeling of local extinction in turbulent

flames," Journal of engineering for gas turbines and power Vol. 118, No. 2, 1996, pp.
292-307.
61. Spalding, D. "Theoretical aspects of flame stabilization: an approximate graphical

method for the flame speed of mixed gases," Aircraft Engineering and Aerospace
Technology Vol. 25, No. 9, 1953, pp. 264-276.
62. Penner, S., and Williams, F. "Recent studies on flame stabilization of premixed
turbulent gases," Applied Mechanics Reviews Vol. 10, No. 6, 1957, pp. 229-237.
63. Zukoski, E. E. "Flame stabilization on bluff bodies at low and intermediate

Reynolds numbers." PhD Thesis, California Institute of Technology, Pasadena, CA,
1954.
152
64. Driscoll, J. F. "Turbulent premixed combustion: Flamelet structure and its effect
on turbulent burning velocities," Progress in Energy and Combustion Science Vol. 34,
No. 1, 2008, pp. 91-134.
65. Dinkelacker, F., Soika, A., Most, D., Hofmann, D., Leipertz, A., Polifke, W., and
Döbbeling, K. "Structure of locally quenched highly turbulent lean premixed flames."
Vol. 27, Elsevier, 1998, pp. 857-865.
66. Ratner, A., Driscoll, J., Donbar, J., Carter, C., and Mullin, J. "Reaction zone
structure of non-premixed turbulent flames in the," Proceedings of the Combustion
Institute Vol. 28, No. 1, 2000, pp. 245-252.
67. Chaudhuri, S., Kostka, S., Renfro, M. W., and Cetegen, B. M. "Blowoff dynamics
of bluff body stabilized turbulent premixed flames," Combustion and Flame Vol. 157,
No. 4, 2010, pp. 790-802.
68. Kariuki, J., Dawson, J. R., and Mastorakos, E. "Measurements in turbulent

premixed bluff body flames close to blow-off," Combustion and Flame, 2012.
69. Muruganandam, T. "Sensing and Dynamics of Lean Blowout in a Swirl Dump

Combustor." PhD Thesis, Georgia Institute of Technology, Atlanta, GA, 2006.
70. Zhang, Q., Shanbhogue, S. J., and Lieuwen, T. "Dynamics of Premixed H2/CH4
Flames Under Near Blowoff Conditions," Journal of engineering for gas turbines and
power Vol. 132, No. 11, 2010, p. 111502.
71. Nair, S. "Acoustic characterization of flame blowout phenomenon." PhD Thesis,

Georgia Institute of Technology, Atlanta, GA, 2006.
72. Lee, J., and Santavicca, D. "Experimental diagnostics for the study of combustion
instabilities in lean premixed combustors," Journal of propulsion and power Vol. 19, No.
5, 2003, pp. 735-750.
73. Docquier, N., and Candel, S. "Combustion control and sensors: a review,"
Progress in Energy and Combustion Science Vol. 28, No. 2, 2002, pp. 107-150.
74. Hill, P. G., and Peterson, C. R. Mechanics and thermodynamics of propulsion:

Addison-Wesley, Reading, Massachusetts, 1992.
75. Spang III, H. A., and Brown, H. "Control of jet engines," Control Engineering
Practice Vol. 7, No. 9, 1999, pp. 1043-1060.
76. May, R. D., Csank, J., Lavelle, T. M., Litt, J. S., and Guo, T. H. "A High-Fidelity
Simulation of a Generic Commercial Aircraft Engine and Controller." NASA/TM-2010-
216810, October, 2010.
153
77. Jaw, L. C., and Mattingly, J. D., Aircraft engine controls, American Institute of
Aeronautics and Astronautics, New York, 2009.
78. Csank, J., May, R. D., Litt, J. S., and Guo, T. H. "Control Design for a Generic
Commercial Aircraft Engine," NASA/TM—2010-216811, October, 2010.
79. Nori, V. N. "Modeling and analysis of chemiluminescence sensing for syngas,

methane and Jet-A combustion." PhD Thesis, Georgia Institute of Technology, Atlanta,
GA, 2008.
80. Walker, J. S., Fast fourier transforms, CRC Press, 1996.
81. Pierce, A. D., Acoustics: An Introduction to its Physical Principles and

Applications, Acoustical Society of America, Melville, NY, 1989.
82. Yates, R. D., and Goodman, D. J., Probability and Stochastic Processes: A
Friendly Introduction for Electrical and Computer Engineers, Jon Wiley & Sons, 2004.
83. Lewis, P. A., and Shedler, G. S. "Simulation of nonhomogeneous Poisson

processes by thinning," Naval Research Logistics Quarterly Vol. 26, No. 3, 1979, pp.
403-413.
84. Li, A. "Improvements to the acoustic pulse reflectometry technique for measuring
duct dimensions." PhD Thesis, The Open University, UK, 2004.
85. Amir, N., Shimony, U., and Rosenhouse, G. "A Discrete Model for Tubular
Acoustic Systems with Varying Cross Section The Direct and Inverse Problems. Part 1:
Theory," Acta Acustica united with Acustica Vol. 81, No. 5, 1995, pp. 450-462.
154
View publication stats

Bompelly Ravi K 201305 PHD

Uploaded by

Copyright:

Available Formats

You might also like

Bompelly Ravi K 201305 PHD

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bompelly Ravi K 201305 PHD

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

LEAN BLOWOUT AND ITS ROBUST SENSING IN SWIRL COMBUSTORS

Thesis · August 2012

The user has requested enhancement of the downloaded file.

Ravi Kiran Bompelly

School of Aerospace Engineering

Dr. Jerry Seitzman, Advisor Dr. Benn T. Zinn

Dr. Timothy Lieuwen Dr. Clarence T. Chang

Dr. Jeff Jagoda

For their love and support

project and providing support and guidance.

Brad Ochs for their help and advice in conducting my experiments.

with them helped me immensely in understanding things clearly and advancing my

knowledge. A special thanks goes to Karthik for helping me with running my

Tom and Caleb, who made the task much easier.

LIST OF TABLES ............................................................................................................. ix

LIST OF FIGURES ............................................................................................................ x

NOMENCLATURE ........................................................................................................ xvi

SUMMARY .................................................................................................................... xvii

CHAPTER 1: INTRODUCTION ....................................................................................... 1

CHAPTER 2: BACKGROUND AND PREVIOUS WORK............................................ 14

CHAPTER 3: EXPERIMENTAL AND MODELING APPROACHES.......................... 32

CHAPTER 4: LBO MARGIN SENSING: OPTICAL DETECTION .............................. 55

CHAPTER 5: LBO MARGIN SENSING: ACOUSTIC DETECTION .......................... 90

CHAPTER 6:LBO MARGIN SENSING: RAPID TRANSIENTS................................ 108

CHAPTER 7: CONCLUSIONS AND RECOMMENDATIONS .................................. 129

APPENDIX B: LDI FUEL NOZZLE TESTING ........................................................... 140

APPENDIX C: LDI COMBUSTOR DEVELOPMENT ................................................ 144

REFERENCES ............................................................................................................... 148

Table 1. Design point specifications of the turbofan engine. ........................................... 50

Table 3. Kolmogorov-Smirnov hypothesis test results. .................................................. 111

Figure 4. Illustration of a swirling flow field structure. .................................................... 18

Figure 5. Illustrative flame configurations in premixed swirl combustors. ...................... 19

Figure 9. Precursor events in the acoustic signal. (Ref. [71]) .......................................... 27

Figure 13. Schematic diagram of the gas fueled combustor setup....................................33

Figure 14. Fuel injection orifice module........................................................................... 34

Figure 17. NASA 9 element LDI injector. [5] ................................................................. 39

Figure 23. Schematic diagram of fuel supply and control setup....................................... 44

Figure 27. Architecture of the engine control system. ...................................................... 51

k Acoustic wave number

Ps 3 Combustor inlet static pressure

U Uniform random number

Wf Fuel flow rate

 Average event occurrence rate

 Expected number of events in a transient

 Normalized equivalence ratio

 Signal standard deviation

LBO Lean Blow Out

LDI Lean Direct Injection

TBE Time Between Events

Lean combustion is increasingly employed in both ground-based gas turbines and

sensors provided a measure of LBO proximity.

precursors in the presence of combustion dynamics and for combustor operating

atmospheric pressure combustor that exhibits pronounced combustion dynamics under a

combustor, operating at elevated pressure and preheat temperature. In the gas-fueled

presence of strong combustion dynamics, and were similar to those observed in

signals have different characteristics under various operating conditions. Low-pass